Models / Changes
The model at that id is not the model you tested.
307 of the 652 open-weight models with readable history (606 tracked) changed a tracked key after their first upload; 137 changed a structural one, 74 of them more than a week after release. Each row is one commit that changed a model's config, tokenizer, chat template, generation defaults or weights, with what that change does.
-46 of the 606 tracked models have no history yet: none of the tracked files (GGUF-only repos, adapters) or files the tracker cannot read (gated). By most consequential class: 210 structural, 242 tokenizer, 212 template, 41 generation-defaults, 0 weights-only, 24 metadata, 12 format-only. 12 rows are format-only re-saves and 2 more carry a format-only key next to a real one. Template changes rendered on the fixed conversations: 149 rendering-changed, 66 cosmetic, 17 the renderer could not run, 0 not rendered yet. 106 rows carry a safety-relevant config change. 392 of the 741 landed more than a week after the model's first commit, so they are not release-day fixes. History reaches back to 2019-02-18; 5 of the changes were caught by a daily snapshot rather than recovered from git. Last read 2026-11-11.
| Date | Model | Class | What changed, and what it does |
|---|---|---|---|
| 2026-10-28 | XiaomiMiMo/MiMo-V2.5 | format-only | format-only re-save, the parameter total the Hub reports moved, with no weight shard recorded as changing; nothing it computes, tokenises or generates changed (2026-10-28). Open in app |
| 2026-10-26 | kindlingai/glm-5.3-exlr8-k2-k3-k4 | format-only | format-only re-save, safetensors files the index does not load were added or changed; every shard a loader reads is identical; nothing it computes, tokenises or generates changed (2026-10-26). Open in app |
| 2026-10-26 | MiniMaxAI/MiniMax-M3-MXFP8 | format-only | format-only re-save, the parameter total the Hub reports moved, with no weight shard recorded as changing; nothing it computes, tokenises or generates changed (2026-10-26). Open in app |
| 2026-10-26 | MiniMaxAI/MiniMax-M2.5 | format-only | format-only re-save, the parameter total the Hub reports moved, with no weight shard recorded as changing; nothing it computes, tokenises or generates changed (2026-10-26). Open in app |
| 2026-10-25 | ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-10-25). Open in app |
| 2026-10-20 | ornith-ai/Ornith-1.5-397B-NVFP4 | format-only | format-only re-save, safetensors files the index does not load were added or changed; every shard a loader reads is identical; nothing it computes, tokenises or generates changed (2026-10-20). Open in app |
| 2026-10-16 | ornith-ai/Ornith-1.5-122B | template rendering-changed | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-10-16). Open in app |
| 2026-10-16 | ornith-ai/Ornith-1.5-122B | template rendering-changed | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-10-16). Open in app |
| 2026-10-16 | ornith-ai/Ornith-1.5-122B | template rendering-changed | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-10-16). Open in app |
| 2026-10-15 | zai-org/GLM-5.3-Flash | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-10-15). Open in app |
| 2026-10-14 | zai-org/GLM-5.3 | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-10-14). Open in app |
| 2026-10-14 | baidu/Unlimited-OCR | structural template | model_type "unlimited-ocr" to "unlimited_ocr", the same name respelt with underscores rather than a different architecture; a chat template was added; apply_chat_template now formats prompts that it previously refused or formatted by default (2026-10-14). Open in app |
| 2026-10-05 | unsloth/gpt-oss-20b-BF16 | tokenizer template rendering-changed default system prompt | EOS stop list [200002,199999] to [200002,199999,200012]; generation now also stops on 200012, and still stops on every token it stopped on before; the chat template changed and renders differently (system+user, multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-10-05).default system prompt: With no system message, the text rendered before the user's message changed: "...uded for every message. Calls to these tools must go to the commentary channel: 'f..." became "...uded for every message. user". Open in app |
| 2026-10-02 | jialinyyzz/humanizer | structural | hidden size 2560 to 3840, MLP width 10240 to 15360, attention heads 8 to 16, layer count 42 to 48, KV heads 2 to 8; derived parameters 4.64B to 11.79B (+7.14B, +14.29 GB of bf16 weights); context window 131072 to 262144 tokens (2026-10-02). Open in app |
| 2026-10-01 | deepseek-ai/DeepSeek-V4.1-Flash | template | a chat template was added; apply_chat_template now formats prompts that it previously refused or formatted by default (2026-10-01). Open in app |
| 2026-09-29 | openbmb/MiniCPM5-2B | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-09-29). Open in app |
| 2026-09-27 | Qdrant/all-MiniLM-L6-v2-onnx | tokenizer | tokenizer model_max_length 512 to 256; truncation defaults change (2026-09-27). Open in app |
| 2026-09-14 | biohub/ESMC-6B | structural tokenizer | EOS declared as 2 (was unset); generation that relied on the library default now stops on this token; config.json BOS declared as 0 (was unset); head dim 64, MLP width 6912, attention heads 40, layer count 80, KV heads 40 declared (each was unset, so the library default for the class applied) (2026-09-14). Open in app |
| 2026-09-05 | CMSManhattan/JiRackUltra_1b | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-09-05).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|Assistant|><think>\n" to "<|Assistant|><think>\n\n</think>\n\n". Open in app |
| 2026-09-04 | zai-org/GLM-5.3-Flash | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-09-04). Open in app |
| 2026-09-04 | zai-org/GLM-5.3 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-09-04). Open in app |
| 2026-09-01 | farbodtavakkoli/OTel-2.0-LLM-31B-IT | structural tokenizer template rendering-changed generation prefix | EOS stop list 106 to [1,106]; generation now also stops on 1, and still stops on every token it stopped on before; derived parameters 30.15B to 30.56B (+418.7M) only because the reconstruction follows the class name; no size key changed; architectures declared as ["Gemma4ForConditionalGeneration"] (was unset) (2026-09-01).generation prefix: With add_generation_prompt, what follows the user's message changed from "<turn|>\n<|turn>model\n<|channel>thought\n<channel|>" to "<turn|>\n<|turn>model\n". Open in app |
| 2026-08-29 | ornith-ai/Ornith-1.5-397B-NVFP4 | structural | quantization_config changed; the stored format of the weights is different (2026-08-29). Open in app |
| 2026-08-29 | unsloth/GLM-5.3-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-29). Open in app |
| 2026-08-28 | RadixArk/Qwen3.8-27B-DSpark | structural tokenizer | EOS changed from 248046 to 248044; anything fine-tuned or served against the old value stops at a different token; config.json BOS 248044 removed; MLP width 10240 to 17408, attention heads 40 to 32 (2026-08-28). Open in app |
| 2026-08-26 | zai-org/GLM-5.3-Flash | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-26). Open in app |
| 2026-08-26 | zai-org/GLM-5.3-Flash | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-26). Open in app |
| 2026-08-24 | ai-sage/Giga-Embeddings-instruct-480M-0826 | structural | context window 32768 to 8192 tokens (2026-08-24). Open in app |
| 2026-08-23 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-08-23).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|><|im_start|>assistant\n<think>\n\n</think>" to "<|im_end|><|im_start|>assistant<think>\n\n</think>". Open in app |
| 2026-08-23 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | tokenizer | config.json eos_token_id 248044 to 248046, but generation_config.json still declares [248046], which generate() reads first, so default generation stops on the same tokens; only code that reads config.eos_token_id directly sees the new value; config.json pad_token_id unset to 248044; the tokenizer's pad token, which training pads with, did not change (2026-08-23). Open in app |
| 2026-08-23 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | tokenizer | EOS stop list [248046,248044] to [248046]; generation no longer stops on 248044 and runs past it (2026-08-23). Open in app |
| 2026-08-22 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | structural tokenizer template rendering-changed tool block generation prefix | derived parameters 26.79B to 26.79B (+7.0M) only because the reconstruction follows the class name; no size key changed; architectures ["Qwen3_5ForCausalLM"] to ["Qwen3_5ForConditionalGeneration"]; a different class is instantiated from the same id; the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-08-22).tool block: A tool list passed to the template reached the prompt before and is dropped now.generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n" to "<|im_end|><|im_start|>assistant\n<think>\n\n</think>". Open in app |
| 2026-08-22 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | structural tokenizer | derived parameters 26.79B to 26.79B (-7.0M) only because the reconstruction follows the class name; no size key changed; architectures ["Qwen3_5ForConditionalGeneration"] to ["Qwen3_5ForCausalLM"]; a different class is instantiated from the same id; add_bos_token false to unset; encoding the same text yields different ids (2026-08-22). Open in app |
| 2026-08-21 | bloomer010/Ling-3.0-tiny-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-21). Open in app |
| 2026-08-20 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | template rendering-changed default system prompt generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-08-20).default system prompt: With no system message, the text rendered before the user's message removed: "system Reasoning effort is set to xhigh. Please think carefully through the task, val..." became "user".generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n<think>\n" to "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n". Open in app |
| 2026-08-19 | OBLITERATUS/Qwen3.8-27B-OBLITERATED | structural tokenizer template | derived parameters 26.79B to 26.79B (+7.0M) only because the reconstruction follows the class name; no size key changed; architectures ["Qwen3_5ForCausalLM"] to ["Qwen3_5ForConditionalGeneration"]; a different class is instantiated from the same id; a chat template was added; apply_chat_template now formats prompts that it previously refused or formatted by default (2026-08-19). Open in app |
| 2026-08-19 | ornith-ai/Ornith-1.5-397B-FP8 | structural | quantization_config changed; the stored format of the weights is different (2026-08-19). Open in app |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-FP8 | structural | quantization_config changed; the stored format of the weights is different (2026-08-19). Open in app |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B | tokenizer | add_bos_token unset to false; encoding the same text yields different ids (2026-08-19). Open in app |
| 2026-08-14 | ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-14). Open in app |
| 2026-08-13 | webAI-Official/TwIL-LM3 | tokenizer | tokenizer model_max_length 131072 to 65536; truncation defaults change; tokenizer_class unset to "PreTrainedTokenizerFast" (2026-08-13). Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B-FP8 | generation-defaults | generation defaults changed (temperature 0.6 to 1); a call that does not pass these explicitly now generates differently (2026-08-13). Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B-FP8 | template rendering-changed default system prompt | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-08-13).default system prompt: With no system message, the text rendered before the user's message added: "user" became "system Reasoning effort is set to xhigh. Please think carefully through the task, val...". Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B | generation-defaults | generation defaults changed (temperature 0.6 to 1); a call that does not pass these explicitly now generates differently (2026-08-13). Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B | template rendering-changed default system prompt | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-08-13).default system prompt: With no system message, the text rendered before the user's message added: "user" became "system Reasoning effort is set to xhigh. Please think carefully through the task, val...". Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B-FP8 | tokenizer | tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-08-13). Open in app |
| 2026-08-13 | Qwen/Qwen3.8-27B | tokenizer | tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-08-13). Open in app |
| 2026-08-10 | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | structural | context window 262144 to 1048576 tokens (2026-08-10). Open in app |
| 2026-08-05 | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-05). Open in app |
| 2026-08-05 | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-08-05). Open in app |
| 2026-08-03 | farbodtavakkoli/OTel-2.0-LLM-31B-IT | tokenizer | EOS changed from "<eos>" to "<turn|>"; anything fine-tuned or served against the old value stops at a different token (2026-08-03). Open in app |
| 2026-08-01 | poolside/Laguna-S-2.1-NVFP4 | structural | context window 262144 to 1048576 tokens; quantization_config changed; the stored format of the weights is different (2026-08-01). Open in app |
| 2026-07-31 | RadixArk/Kimi-K3-DSpark | structural | rope_scaling added ({"factor":16,"original_max_position_embeddings":65536,"ro...); positions are now scaled (2026-07-31). Open in app |
| 2026-07-29 | poolside/Laguna-XS-2.1 | generation-defaults | generation defaults changed (top_k unset to 20); a call that does not pass these explicitly now generates differently (2026-07-29). Open in app |
| 2026-07-25 | farbodtavakkoli/OTel-2.0-LLM-31B-IT | generation-defaults | generation_config.json was removed; whatever defaults it set now come from the library (2026-07-25). Open in app |
| 2026-07-24 | farbodtavakkoli/OTel-2.0-LLM-31B-IT | structural tokenizer template cosmetic | config.json eos_token_id [1,106] to 1, but generation_config.json still declares [1,106,50], which generate() reads first, so default generation stops on the same tokens; only code that reads config.eos_token_id directly sees the new value; derived parameters 30.56B to 30.15B (-418.7M) only because the reconstruction follows the class name; no size key changed; architectures ["Gemma4ForConditionalGeneration"] removed; the class is now inferred from model_type (2026-07-24). Open in app |
| 2026-07-24 | k-chirkunov/gemma4-e4b-claims-comparison | tokenizer template cosmetic | pad token "<eos>" to "<pad>"; batched training masks a different id; the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-07-24). Open in app |
| 2026-07-23 | ibm-research/MoLFormer-XL-both-10pct | tokenizer format-only | EOS declared as "<eos>" (was unset); generation that relied on the library default now stops on this token; BOS declared as "<bos>" (was unset); tokenizer_class "MolformerTokenizer" to "PreTrainedTokenizerFast" (2026-07-23). Open in app |
| 2026-07-23 | AnkitAI/Parable-Qwen3-4B-Claude-Fable-5-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-07-23). Open in app |
| 2026-07-22 | poolside/Laguna-S-2.1-NVFP4 | structural | quantization_config changed; the stored format of the weights is different (2026-07-22). Open in app |
| 2026-07-21 | poolside/Laguna-S-2.1-NVFP4 | template generation-defaults rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering; generation defaults changed (max_new_tokens 32768 to unset); a call that does not pass these explicitly now generates differently (2026-07-21).generation prefix: With add_generation_prompt, what follows the user's message changed from "</user>\n<assistant></think>" to "</user>\n<assistant><think>". Open in app |
| 2026-07-21 | poolside/Laguna-S-2.1 | template generation-defaults rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering; generation defaults changed (max_new_tokens 32768 to unset); a call that does not pass these explicitly now generates differently (2026-07-21).generation prefix: With add_generation_prompt, what follows the user's message changed from "</user>\n<assistant></think>" to "</user>\n<assistant><think>". Open in app |
| 2026-07-21 | poolside/Laguna-S-2.1 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-07-21). Open in app |
| 2026-07-20 | tencent/Hy3 | declared dtype unset to "bfloat16" (2026-07-20). Open in app | |
| 2026-07-18 | unsloth/Ornith-1.0-9B-GGUF | template rendering-changed | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-07-18). Open in app |
| 2026-07-15 | thinkingmachines/Inkling | template | a chat template was added; apply_chat_template now formats prompts that it previously refused or formatted by default (2026-07-15). Open in app |
| 2026-07-15 | google/gemma-4-31B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-07-15). Open in app |
| 2026-07-15 | google/gemma-4-26B-A4B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-07-15). Open in app |
| 2026-07-14 | microsoft/phi-4 | tokenizer | EOS stop list 100265 to [100257,100265]; generation now also stops on 100257, and still stops on every token it stopped on before (2026-07-14). Open in app |
| 2026-07-13 | ibm-granite/granite-vision-4.1-4b | structural | vocab_size changed from 100353 to 100352; the embedding and output head change shape, so adapters and merges made against the old vocabulary no longer load cleanly; derived parameters 3.32B to 3.32B (-2.6K, -0.00 GB of bf16 weights) (2026-07-13). Open in app |
| 2026-07-12 | ibm-granite/granite-vision-4.1-4b | tokenizer | tokenizer_class "GPT2Tokenizer" to "PreTrainedTokenizerFast" (2026-07-12). Open in app |
| 2026-07-08 | OpenMOSS-Team/MOSS-Transcribe-Diarize | tokenizer | add_bos_token false to unset; encoding the same text yields different ids (2026-07-08). Open in app |
| 2026-07-07 | poolside/Laguna-XS-2.1 | template rendering-changed default system prompt generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-07-07).default system prompt: With no system message, the text rendered before the user's message added: "" became "You are a helpful, conversationally-fluent assistant made by Poolside. You are here t...".generation prefix: With add_generation_prompt, what follows the user's message changed from "\n</user>\n<assistant>\n</think>" to "</user>\n<assistant></think>". Open in app |
| 2026-07-06 | OpenMOSS-Team/MOSS-Transcribe-Diarize | structural | context window 40960 to 131072 tokens; declared dtype unset to "bfloat16" (2026-07-06). Open in app |
| 2026-07-05 | tencent/Hy3 | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-07-05). Open in app |
| 2026-07-03 | Bahushruth/Qwen3.6-35B-A3B-abliterated-v4 | structural | architectures ["Qwen3_5MoeForConditionalGeneration"] to ["Qwen3_5MoeForCausalLM"]; a different class is instantiated from the same id (2026-07-03). Open in app |
| 2026-07-03 | Bahushruth/Qwen3.6-35B-A3B-abliterated-v4 | structural | architectures ["Qwen3_5MoeForCausalLM"] to ["Qwen3_5MoeForConditionalGeneration"]; a different class is instantiated from the same id (2026-07-03). Open in app |
| 2026-06-30 | nvidia/Qwen3.6-27B-NVFP4 | structural | quantization_config changed; the stored format of the weights is different (2026-06-30). Open in app |
| 2026-06-27 | google/tipsv2-so400m14 | structural tokenizer | hidden size 1152, MLP width 4304, attention heads 16, layer count 27 declared (each was unset, so the library default for the class applied); context window unset to 64 tokens; config.json pad_token_id unset to 0; the tokenizer's pad token, which training pads with, did not change (2026-06-27). Open in app |
| 2026-06-27 | google/tipsv2-b14 | structural tokenizer | hidden size 768, MLP width 3072, attention heads 12, layer count 12 declared (each was unset, so the library default for the class applied); context window unset to 64 tokens; config.json pad_token_id unset to 0; the tokenizer's pad token, which training pads with, did not change (2026-06-27). Open in app |
| 2026-06-25 | ornith-ai/Ornith-1.0-397B-FP8 | template rendering-changed | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-06-25). Open in app |
| 2026-06-24 | nvidia/MiniMax-M3-NVFP4 | tokenizer | EOS declared as "[e~[" (was unset); generation that relied on the library default now stops on this token; BOS declared as "]~b]" (was unset); pad token unset to "]!p~["; batched training masks a different id (2026-06-24). Open in app |
| 2026-06-19 | z-lab/Qwen3.6-35B-A3B-DFlash | structural tokenizer | EOS changed from 248046 to 248044; anything fine-tuned or served against the old value stops at a different token; layer count 8 to 6, KV heads 4 to 8; derived parameters 1.47B to 1.37B (-100.7M, -0.20 GB of bf16 weights) (2026-06-19). Open in app |
| 2026-06-15 | moonshotai/Kimi-K2.7-Code | template generation-defaults cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; generation defaults changed (temperature unset to 1, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2026-06-15). Open in app |
| 2026-05-27 | poolside/Laguna-XS.2 | generation-defaults | generation defaults changed (max_new_tokens 2048 to 4096, min_p unset to 0, temperature 0.7 to 1, top_p 0.9 to 1); a call that does not pass these explicitly now generates differently (2026-05-27). Open in app |
| 2026-05-26 | poolside/Laguna-XS.2 | structural | context window 131072 to 262144 tokens (2026-05-26). Open in app |
| 2026-05-18 | google/gemma-4-31B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-18). Open in app |
| 2026-05-18 | google/gemma-4-26B-A4B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-18). Open in app |
| 2026-05-12 | moonshotai/Kimi-K2.6 | format-only | format-only re-save, tokenizer class renamed between its Fast and plain spelling (one class per model since transformers v5); nothing it computes, tokenises or generates changed (2026-05-12). Open in app |
| 2026-05-11 | moonshotai/Kimi-K2.6 | format-only | format-only re-save, tokenizer class renamed between its Fast and plain spelling (one class per model since transformers v5); nothing it computes, tokenises or generates changed (2026-05-11). Open in app |
| 2026-05-08 | XiaomiMiMo/MiMo-V2.5 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-08). Open in app |
| 2026-05-07 | XiaomiMiMo/MiMo-V2.5 | tokenizer | EOS changed from [151643,151645,1561672] to [151643,151645,151672]; anything fine-tuned or served against the old value stops at a different token (2026-05-07). Open in app |
| 2026-05-07 | nvidia/Gemma-4-26B-A4B-NVFP4 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-07). Open in app |
| 2026-05-05 | zeroentropy/zerank-2-reranker | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-05). Open in app |
| 2026-05-01 | nvidia/Gemma-4-31B-IT-NVFP4 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-05-01). Open in app |
| 2026-04-30 | moonshotai/Kimi-K2.5 | structural | quantization_config changed; the stored format of the weights is different (2026-04-30). Open in app |
| 2026-04-30 | moonshotai/Kimi-K2.6 | structural | quantization_config changed; the stored format of the weights is different (2026-04-30). Open in app |
| 2026-04-29 | sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-29). Open in app |
| 2026-04-28 | ibm-granite/granite-vision-4.1-4b | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-04-28). Open in app |
| 2026-04-28 | google/gemma-4-31B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-04-28). Open in app |
| 2026-04-28 | google/gemma-4-26B-A4B-it | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-04-28). Open in app |
| 2026-04-28 | XiaomiMiMo/MiMo-V2.5 | structural template rendering-changed generation prefix | context window 262144 to 1048576 tokens; rope_theta 5000000 to 10000000; every position embedding changes; the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-28).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|><|im_start|>assistant\n<think></think>" to "<|im_end|><|im_start|>assistant\n". Open in app |
| 2026-04-28 | nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 | tokenizer | pad token unset to "<|im_end|>"; batched training masks a different id (2026-04-28). Open in app |
| 2026-04-27 | XiaomiMiMo/MiMo-V2.5 | tokenizer template generation-defaults rendering-changed default system prompt tool block generation prefix | EOS stop list [151643,151645] to [151643,151645,1561672]; generation now also stops on 1561672, and still stops on every token it stopped on before; the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering; generation defaults changed (temperature unset to 1, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2026-04-27).default system prompt: With no system message, the text rendered before the user's message changed: "system You are a helpful assistant. user" became "system You are MiMo, a helpful AI assistant engineered by Xiaomi. user".tool block: A tool list passed to the template is rendered into the prompt now and was dropped before.generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n" to "<|im_end|><|im_start|>assistant\n<think></think>". Open in app |
| 2026-04-22 | Qwen/Qwen3.6-27B | tokenizer | tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-04-22). Open in app |
| 2026-04-22 | Qwen/Qwen3.6-27B | generation-defaults | generation defaults changed (temperature 0.6 to 1); a call that does not pass these explicitly now generates differently (2026-04-22). Open in app |
| 2026-04-21 | nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-21).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n\n<|im_start|>assistant\n<think>\n" to "<|im_end|>\n<|im_start|>assistant\n<think>\n". Open in app |
| 2026-04-20 | nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 | generation-defaults | generation defaults changed (do_sample unset to true, max_new_tokens unset to 16384, repetition_penalty unset to 1, temperature unset to 0.6); a call that does not pass these explicitly now generates differently (2026-04-20). Open in app |
| 2026-04-20 | nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 | structural | architectures ["Nemotron_3_Nano_Omni_30B_A3B_Reasoning"] to ["NemotronH_Nano_Omni_Reasoning_V3"]; a different class is instantiated from the same id (2026-04-20). Open in app |
| 2026-04-20 | MiniMaxAI/MiniMax-M2.7 | structural | context window 196608 to 204800 tokens (2026-04-20). Open in app |
| 2026-04-18 | nvidia/Gemma-4-31B-IT-NVFP4 | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-18). Open in app |
| 2026-04-17 | z-lab/Qwen3.6-35B-A3B-DFlash | structural | rope_scaling rewritten in a different format ({"beta_fast":32,"beta_slow":1,"factor":64,"original_max_p... to {"beta_fast":32,"beta_slow":1,"factor":64,"original_max_p...); a loader that only reads the old format rejects or ignores the new one (2026-04-17). Open in app |
| 2026-04-16 | Qwen/Qwen3-Reranker-8B | template rendering-changed tool block | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-16).tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2026-04-16 | Qwen/Qwen3-Reranker-4B | template rendering-changed tool block | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-16).tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2026-04-16 | Qwen/Qwen3-Reranker-0.6B | template rendering-changed tool block | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-16).tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2026-04-16 | zai-org/GLM-5.1 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-04-16). Open in app |
| 2026-04-10 | nvidia/llama-nemotron-rerank-1b-v2 | format-only | format-only re-save, RoPE settings rewritten in the transformers v5 rope_parameters form with the same values; nothing it computes, tokenises or generates changed (2026-04-10). Open in app |
| 2026-04-10 | google/gemma-4-31B-it | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-10). Open in app |
| 2026-04-10 | google/gemma-4-26B-A4B-it | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-10). Open in app |
| 2026-04-09 | MiniMaxAI/MiniMax-M2.7 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-09).default system prompt: With no system message, the text rendered before the user's message changed: "...are a helpful assistant.[e~[ ]~b]user" became "...are a helpful assistant. Your name is MiniMax-M2.7 and is built by MiniMax.[e~[ ]~...". Open in app |
| 2026-04-09 | MiniMaxAI/MiniMax-M2.7 | declared dtype unset to "bfloat16" (2026-04-09). Open in app | |
| 2026-04-09 | jinaai/jina-reranker-m0 | template rendering-changed | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-04-09). Open in app |
| 2026-04-08 | mixedbread-ai/mxbai-rerank-large-v2 | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2026-04-08). Open in app |
| 2026-04-07 | nomic-ai/nomic-embed-text-v1.5 | structural tokenizer | head dim 64 declared (it was unset, so the library default for the class applied); our reconstruction builds the same graph from both configs (same parameter count and structure); context window 8192 to 2048 tokens; rope_theta unset to 1000; every position embedding changes (2026-04-07). Open in app |
| 2026-04-07 | nomic-ai/nomic-embed-text-v1 | structural tokenizer | head dim 64 declared (it was unset, so the library default for the class applied); our reconstruction builds the same graph from both configs (same parameter count and structure); rope_scaling added ({"factor":2,"rope_type":"dynamic"}); positions are now scaled; rope_theta unset to 1000; every position embedding changes (2026-04-07). Open in app |
| 2026-04-02 | nvidia/Gemma-4-31B-IT-NVFP4 | tokenizer | pad token "<eos>" to "<pad>"; batched training masks a different id (2026-04-02). Open in app |
| 2026-03-30 | nvidia/llama-nemotron-embed-1b-v2 | format-only | format-only re-save, RoPE settings rewritten in the transformers v5 rope_parameters form with the same values; nothing it computes, tokenises or generates changed (2026-03-30). Open in app |
| 2026-03-16 | nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 | generation-defaults | generation defaults changed (temperature unset to 1, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2026-03-16). Open in app |
| 2026-03-11 | zai-org/GLM-5 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-03-11). Open in app |
| 2026-03-09 | stepfun-ai/Step-3.5-Flash | structural | derived parameters 8.28B to 8.81B only because our reconstruction stops falling back to its own default; whether the loaded model changed depends on the library default, which this record does not check; tie_word_embeddings unset to false; the output head is no longer shared with the embedding (2026-03-09). Open in app |
| 2026-03-06 | sarvamai/sarvam-30b | structural | context window 32768 to 131072 tokens (2026-03-06). Open in app |
| 2026-03-04 | nvidia/Kimi-K2.5-NVFP4 | structural | quantization_config changed; the stored format of the weights is different (2026-03-04). Open in app |
| 2026-03-04 | sarvamai/sarvam-30b | structural | context window 131072 to 32768 tokens (2026-03-04). Open in app |
| 2026-03-02 | unsloth/Qwen3.5-9B-GGUF | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-03-02).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n<think>\n" to "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n". Open in app |
| 2026-03-01 | Qwen/Qwen3.5-0.8B | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-03-01).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n<think>\n" to "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n". Open in app |
| 2026-03-01 | Qwen/Qwen3.5-2B | template rendering-changed generation prefix | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-03-01).generation prefix: With add_generation_prompt, what follows the user's message changed from "<|im_end|>\n<|im_start|>assistant\n<think>\n" to "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n". Open in app |
| 2026-03-01 | Qwen/Qwen3.5-0.8B | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-03-01). Open in app |
| 2026-03-01 | Qwen/Qwen3.5-2B | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-03-01). Open in app |
| 2026-03-01 | Qwen/Qwen3.5-9B | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-03-01). Open in app |
| 2026-03-01 | Qwen/Qwen3.5-4B | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-03-01). Open in app |
| 2026-02-27 | amd/DeepSeek-R1-MXFP4 | structural | quantization_config changed; the stored format of the weights is different (2026-02-27). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-27B | tokenizer | tokenizer_class "Qwen3_5Tokenizer" to "Qwen2Tokenizer" (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-35B-A3B | tokenizer | tokenizer_class "Qwen3_5Tokenizer" to "Qwen2Tokenizer" (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-27B | tokenizer | tokenizer_class "Qwen2Tokenizer" to "Qwen3_5Tokenizer" (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-35B-A3B | tokenizer | tokenizer_class "Qwen2Tokenizer" to "Qwen3_5Tokenizer" (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-27B | tokenizer | tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-27B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-35B-A3B | tokenizer | tokenizer model_max_length 131072 to 262144; truncation defaults change (2026-02-24). Open in app |
| 2026-02-24 | Qwen/Qwen3.5-35B-A3B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2026-02-24). Open in app |
| 2026-02-19 | jinaai/jina-embeddings-v5-text-nano-retrieval | tokenizer | tokenizer_class "TokenizersBackend" to "PreTrainedTokenizer" (2026-02-19). Open in app |
| 2026-02-16 | jinaai/jina-embeddings-v5-text-nano-retrieval | format-only | format-only re-save, RoPE settings rewritten in the transformers v5 rope_parameters form with the same values; tokenizer class renamed between its Fast and plain spelling (one class per model since transformers v5); nothing it computes, tokenises or generates changed (2026-02-16). Open in app |
| 2026-02-16 | jinaai/jina-embeddings-v5-text-nano-retrieval | structural | model_type "jina_embeddings_v5" to "eurobert"; a different class is instantiated from the same id (2026-02-16). Open in app |
| 2026-02-12 | MiniMaxAI/MiniMax-M2.5 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-02-12).default system prompt: With no system message, the text rendered before the user's message changed: "...are a helpful assistant.[e~[ ]~b]user" became "...are a helpful assistant. Your name is MiniMax-M2.5 and is built by MiniMax.[e~[ ]~...". Open in app |
| 2026-02-11 | jinaai/jina-embeddings-v5-text-nano-retrieval | structural | architectures ["EuroBertForMaskedLM"] to ["EuroBertModel"]; a different class is instantiated from the same id (2026-02-11). Open in app |
| 2026-02-10 | jinaai/jina-embeddings-v5-text-nano-retrieval | structural | architectures ["EuroBertModel"] to ["EuroBertForMaskedLM"]; a different class is instantiated from the same id (2026-02-10). Open in app |
| 2026-02-02 | jinaai/jina-embeddings-v5-text-small | structural | derived parameters 596.0M to 537.4M (-58.6M) only because the reconstruction follows the class name; no size key changed; context window 40960 to 32768 tokens; model_type "qwen3" to "jina_embeddings_v5"; a different class is instantiated from the same id (2026-02-02). Open in app |
| 2026-02-02 | jinaai/jina-embeddings-v5-text-nano | structural | rope_theta 250000 to 1000000; every position embedding changes (2026-02-02). Open in app |
| 2026-02-02 | stepfun-ai/Step-3.5-Flash | generation-defaults | generation_config.json was removed; whatever defaults it set now come from the library (2026-02-02). Open in app |
| 2026-01-30 | unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-01-30). Open in app |
| 2026-01-29 | unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-01-29). Open in app |
| 2026-01-29 | moonshotai/Kimi-K2.5 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2026-01-29).default system prompt: With no system message, the text rendered before the user's message removed: "system You are Kimi, an AI assistant created by Moonshot AI. user" became "user". Open in app |
| 2026-01-28 | sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 | tokenizer | tokenizer_class unset to "PreTrainedTokenizerFast" (2026-01-28). Open in app |
| 2025-12-15 | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-12-15). Open in app |
| 2025-12-15 | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-12-15). Open in app |
| 2025-12-10 | allenai/Olmo-3-7B-Think | generation-defaults | generation defaults changed (do_sample unset to true); a call that does not pass these explicitly now generates differently (2025-12-10). Open in app |
| 2025-12-08 | MiniMaxAI/MiniMax-M2 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-12-08).default system prompt: With no system message, the text rendered before the user's message changed: "]~b]system You are a helpful assistant.[e~[ ]~b]user" became "]~b]system You are MiniMax-M2, a helpful AI assistant built by MiniMax. Knowledge cut...". Open in app |
| 2025-12-06 | nvidia/Qwen2.5-VL-7B-Instruct-NVFP4 | structural | quantization_config changed; the stored format of the weights is different (2025-12-06). Open in app |
| 2025-12-05 | cjvt/GaMS3-12B-Instruct | generation-defaults | generation defaults changed (temperature 0.2 to 0.6, top_k unset to 64, top_p 0.7 to 0.95); a call that does not pass these explicitly now generates differently (2025-12-05). Open in app |
| 2025-12-05 | cjvt/GaMS3-12B-Instruct | generation-defaults | generation_config.json is still in the repository but is no longer valid JSON, so none of its settings can be read from it (2025-12-05). Open in app |
| 2025-12-03 | Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 | structural tokenizer | config.json BOS declared as 151643 (was unset); MLP width 5472 to 6144; our reconstruction builds the same graph from both configs (same parameter count and structure); quantization_config changed; the stored format of the weights is different (2025-12-03). Open in app |
| 2025-12-03 | Qwen/Qwen3-Coder-30B-A3B-Instruct | structural tokenizer | config.json BOS declared as 151643 (was unset); MLP width 5472 to 6144; our reconstruction builds the same graph from both configs (same parameter count and structure) (2025-12-03). Open in app |
| 2025-12-01 | deepseek-ai/DeepSeek-V3.2 | generation-defaults | generation defaults changed (temperature 0.6 to 1); a call that does not pass these explicitly now generates differently (2025-12-01). Open in app |
| 2025-11-26 | allenai/Olmo-3-7B-Instruct | generation-defaults | generation defaults changed (do_sample unset to true); a call that does not pass these explicitly now generates differently (2025-11-26). Open in app |
| 2025-11-21 | allenai/Olmo-3-7B-Instruct | generation-defaults | generation defaults changed (max_new_tokens unset to 32768, temperature unset to 0.6, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2025-11-21). Open in app |
| 2025-11-21 | allenai/Olmo-3-7B-Think | generation-defaults | generation defaults changed (max_new_tokens unset to 32768, temperature unset to 0.6, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2025-11-21). Open in app |
| 2025-11-14 | swiss-ai/Apertus-8B-Instruct-2509 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-11-14). Open in app |
| 2025-11-07 | MiniMaxAI/MiniMax-M2 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-11-07). Open in app |
| 2025-11-06 | MiniMaxAI/MiniMax-M2 | structural tokenizer | EOS declared as 200020 (was unset); generation that relied on the library default now stops on this token; generation_config.json BOS declared as 200019 (was unset); model_type "minimax" to "minimax_m2"; a different class is instantiated from the same id (2025-11-06). Open in app |
| 2025-10-31 | allenai/Olmo-3-7B-Think | structural | architectures ["Olmo2RetrofitForCausalLM"] to ["Olmo3ForCausalLM"]; a different class is instantiated from the same id (2025-10-31). Open in app |
| 2025-10-27 | MiniMaxAI/MiniMax-M2 | structural | model_type "mixtral" to "minimax"; a different class is instantiated from the same id (2025-10-27). Open in app |
| 2025-10-27 | MiniMaxAI/MiniMax-M2 | generation-defaults | generation defaults changed (top_k unset to 40); a call that does not pass these explicitly now generates differently (2025-10-27). Open in app |
| 2025-10-25 | MiniMaxAI/MiniMax-M2 | structural | architectures ["MiniMaxM2MiniForCausalLM"] to ["MiniMaxM2ForCausalLM"]; a different class is instantiated from the same id (2025-10-25). Open in app |
| 2025-10-22 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-10-22). Open in app |
| 2025-10-22 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-10-22). Open in app |
| 2025-10-10 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-10-10). Open in app |
| 2025-10-08 | nvidia/Qwen2.5-VL-7B-Instruct-NVFP4 | format-only | format-only re-save, RoPE settings rewritten in the transformers v5 rope_parameters form with the same values; nothing it computes, tokenises or generates changed (2025-10-08). Open in app |
| 2025-10-04 | amd/DeepSeek-R1-MXFP4 | structural | quantization_config changed; the stored format of the weights is different (2025-10-04). Open in app |
| 2025-09-23 | ai-sage/Giga-Embeddings-instruct | structural tokenizer | model_type "gigar" to "gigarembed"; a different class is instantiated from the same id; declared dtype "bfloat16" to "float32"; tokenizer_class "PreTrainedTokenizerFast" to "PreTrainedTokenizer" (2025-09-23). Open in app |
| 2025-09-23 | ai-sage/Giga-Embeddings-instruct | structural template | head dim 128 to 64, KV heads 2 to 16; derived parameters 3.30B to 2.50B (-801.0M, -1.60 GB of bf16 weights); rope_theta 1300 to 100000; every position embedding changes (2025-09-23). Open in app |
| 2025-09-22 | PrimeIntellect/Qwen3-0.6B | structural template rendering-changed default system prompt | declared dtype "float32" to "bfloat16"; the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-22).default system prompt: With no system message, the text rendered before the user's message removed: "system You are Qwen, created by Alibaba Cloud. You are a helpful assistant. user" became "user". Open in app |
| 2025-09-17 | Qwen/Qwen3-4B-Instruct-2507-FP8 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 262144 to 1010000; truncation defaults change (2025-09-17). Open in app |
| 2025-09-17 | Qwen/Qwen3-4B-Instruct-2507 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 262144 to 1010000; truncation defaults change (2025-09-17). Open in app |
| 2025-09-17 | Qwen/Qwen3-30B-A3B-Instruct-2507 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-09-17). Open in app |
| 2025-09-17 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 262144 to 1010000; truncation defaults change (2025-09-17). Open in app |
| 2025-09-17 | Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 262144 to 1010000; truncation defaults change (2025-09-17). Open in app |
| 2025-09-11 | OpenGVLab/InternVL3-8B | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message added: "user" became "system 你是书生·万象,英文名是InternVL,是由上海人工智能实验室、清华大学及多家合作单位联合开发的多模态大语言模型。 user". Open in app |
| 2025-09-11 | OpenGVLab/InternVL3-1B | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message added: "user" became "system 你是书生·万象,英文名是InternVL,是由上海人工智能实验室、清华大学及多家合作单位联合开发的多模态大语言模型。 user". Open in app |
| 2025-09-11 | OpenGVLab/InternVL2_5-4B | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message added: "user" became "system 你是书生·万象,英文名是InternVL,是由上海人工智能实验室、清华大学及多家合作单位联合开发的多模态大语言模型。 user". Open in app |
| 2025-09-11 | OpenGVLab/InternVL2_5-4B | template rendering-changed default system prompt tool block | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message removed: "system You are Qwen, created by Alibaba Cloud. You are a helpful assistant. user" became "user".tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2025-09-11 | OpenGVLab/InternVL3-1B | template rendering-changed default system prompt tool block | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message removed: "system You are Qwen, created by Alibaba Cloud. You are a helpful assistant. user" became "user".tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2025-09-11 | OpenGVLab/InternVL3-8B | template rendering-changed default system prompt tool block | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-11).default system prompt: With no system message, the text rendered before the user's message removed: "system You are Qwen, created by Alibaba Cloud. You are a helpful assistant. user" became "user".tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2025-09-01 | swiss-ai/Apertus-8B-Instruct-2509 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-01).default system prompt: With no system message, the text rendered before the user's message changed: "Deliberation: disabled T..." became "You are Apertus, a helpful assistant created by the SwissAI initiative. Knowledge cut...". Open in app |
| 2025-09-01 | swiss-ai/Apertus-8B-Instruct-2509 | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-09-01).default system prompt: With no system message, the text rendered before the user's message changed: "You are Apertus, a helpful assistant created by the SwissAI initiative. Knowledge cut..." became "Deliberation: disabled T...". Open in app |
| 2025-08-26 | deepseek-ai/DeepSeek-V3.1 | structural | quantization_config changed; the stored format of the weights is different (2025-08-26). Open in app |
| 2025-08-21 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | tokenizer | tokenizer model_max_length 1e+30 to 131072; truncation defaults change (2025-08-21). Open in app |
| 2025-08-21 | Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-21). Open in app |
| 2025-08-21 | Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-21). Open in app |
| 2025-08-21 | Qwen/Qwen3-Coder-30B-A3B-Instruct | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-21). Open in app |
| 2025-08-21 | Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2025-08-21). Open in app |
| 2025-08-21 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | tokenizer | config.json eos_token_id 2 to 12, but generation_config.json still declares [2,11,12], which generate() reads first, so default generation stops on the same tokens; only code that reads config.eos_token_id directly sees the new value (2025-08-21). Open in app |
| 2025-08-20 | mixedbread-ai/mxbai-rerank-large-v2 | tokenizer | config.json pad_token_id unset to 151645; the tokenizer's pad token, which training pads with, did not change (2025-08-20). Open in app |
| 2025-08-15 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | tokenizer template rendering-changed tool block | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering; add_bos_token true to false; encoding the same text yields different ids (2025-08-15).tool block: A tool list passed to the template is rendered into the prompt now and was dropped before. Open in app |
| 2025-08-13 | openai/gpt-oss-120b | tokenizer stop tokens | EOS stop list [200002,199999] to [200002,199999,200012]; generation now also stops on 200012, and still stops on every token it stopped on before (2025-08-13).stop tokens: generation_config now stops on <|call|>, which ends a tool call in the template; before, nothing in its stop list did. Open in app |
| 2025-08-13 | openai/gpt-oss-20b | tokenizer stop tokens | EOS stop list [200002,199999] to [200002,199999,200012]; generation now also stops on 200012, and still stops on every token it stopped on before (2025-08-13).stop tokens: generation_config now stops on <|call|>, which ends a tool call in the template; before, nothing in its stop list did. Open in app |
| 2025-08-11 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (system+user, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-11). Open in app |
| 2025-08-11 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (system+user, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-11). Open in app |
| 2025-08-08 | openai/gpt-oss-20b | template rendering-changed | the chat template changed and renders differently (system+user, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-08). Open in app |
| 2025-08-08 | openai/gpt-oss-120b | template rendering-changed | the chat template changed and renders differently (system+user, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-08). Open in app |
| 2025-08-08 | Qwen/Qwen3-30B-A3B-Instruct-2507 | tokenizer | tokenizer model_max_length 262144 to 1010000; truncation defaults change (2025-08-08). Open in app |
| 2025-08-07 | openai/gpt-oss-120b | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-07). Open in app |
| 2025-08-07 | openai/gpt-oss-20b | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-07). Open in app |
| 2025-08-07 | Qwen/Qwen3-Coder-30B-A3B-Instruct | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2025-08-07). Open in app |
| 2025-08-07 | Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2025-08-07). Open in app |
| 2025-08-06 | unsloth/gpt-oss-20b-GGUF | structural | declared dtype unset to "bfloat16"; quantization_config changed; the stored format of the weights is different (2025-08-06). Open in app |
| 2025-08-05 | unsloth/gpt-oss-20b-BF16 | template rendering-changed default system prompt | the chat template changed and renders differently (system+user, multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-05).default system prompt: With no system message, the text rendered before the user's message changed: "...uded for every message. user" became "...uded for every message. Calls to these tools must go to the commentary channel: 'f...". Open in app |
| 2025-08-05 | openai/gpt-oss-120b | template rendering-changed default system prompt | the chat template changed and renders differently (system+user, multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-05).default system prompt: With no system message, the text rendered before the user's message changed: "...uded for every message. Calls to these tools must go to the commentary channel: 'f..." became "...uded for every message. user". Open in app |
| 2025-08-05 | openai/gpt-oss-20b | template rendering-changed default system prompt | the chat template changed and renders differently (system+user, multi-turn); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-08-05).default system prompt: With no system message, the text rendered before the user's message changed: "...uded for every message. Calls to these tools must go to the commentary channel: 'f..." became "...uded for every message. user". Open in app |
| 2025-08-05 | unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-08-05). Open in app |
| 2025-07-31 | Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json pad_token_id unset to 151643; generate() fills finished sequences in a batch with this id, while training pads with the tokenizer's pad token, which did not change; generation defaults changed (do_sample unset to true, repetition_penalty unset to 1.05, temperature unset to 0.7, top_k unset to 20); a call that does not pass these explicitly now generates differently (2025-07-31). Open in app |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | generation-defaults | generation defaults changed (temperature 0.6 to 0.7, top_p 0.95 to 0.8); a call that does not pass these explicitly now generates differently (2025-07-29). Open in app |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2025-07-29). Open in app |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2025-07-29). Open in app |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | generation-defaults | generation defaults changed (temperature 0.7 to 0.6, top_p 0.8 to 0.95); a call that does not pass these explicitly now generates differently (2025-07-29). Open in app |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json pad_token_id unset to 151643; generate() fills finished sequences in a batch with this id, while training pads with the tokenizer's pad token, which did not change; generation defaults changed (do_sample unset to true, temperature unset to 0.7, top_k unset to 20, top_p unset to 0.8); a call that does not pass these explicitly now generates differently (2025-07-29). Open in app |
| 2025-07-28 | HuggingFaceTB/SmolLM3-3B | structural | tie_word_embeddings unset to true; the output head is now shared with the embedding (2025-07-28). Open in app |
| 2025-07-28 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | structural | MLP width 5504 to 6144; our reconstruction builds the same graph from both configs (same parameter count and structure); context window 1048576 to 262144 tokens; quantization_config changed; the stored format of the weights is different (2025-07-28). Open in app |
| 2025-07-28 | Qwen/Qwen3-30B-A3B-Instruct-2507 | structural | MLP width 5504 to 6144; our reconstruction builds the same graph from both configs (same parameter count and structure); context window 1048576 to 262144 tokens; architectures ["Qwen2MoeForCausalLM"] to ["Qwen3MoeForCausalLM"]; a different class is instantiated from the same id (2025-07-28). Open in app |
| 2025-07-22 | Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS 151643 removed; generation_config.json pad_token_id unset to 151643; generate() fills finished sequences in a batch with this id, while training pads with the tokenizer's pad token, which did not change (2025-07-22). Open in app |
| 2025-07-22 | hmellor/Ilama-3.2-1B | structural | architectures ["LlamaForCausalLM"] to ["IlamaForCausalLM"]; a different class is instantiated from the same id (2025-07-22). Open in app |
| 2025-07-21 | Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json pad_token_id unset to 151643; generate() fills finished sequences in a batch with this id, while training pads with the tokenizer's pad token, which did not change; generation defaults changed (do_sample unset to true, temperature unset to 0.7, top_k unset to 20, top_p unset to 0.8); a call that does not pass these explicitly now generates differently (2025-07-21). Open in app |
| 2025-07-21 | Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | tokenizer template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered; tokenizer model_max_length 131072 to 262144; truncation defaults change (2025-07-21). Open in app |
| 2025-07-21 | Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | structural | MLP width 10944 to 12288; our reconstruction builds the same graph from both configs (same parameter count and structure); context window 1048576 to 262144 tokens; quantization_config changed; the stored format of the weights is different (2025-07-21). Open in app |
| 2025-07-18 | moonshotai/Kimi-K2-Instruct | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-07-18).default system prompt: With no system message, the text rendered before the user's message changed: "system You are a helpful assistant user" became "system You are Kimi, an AI assistant created by Moonshot AI. user". Open in app |
| 2025-07-15 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-07-15). Open in app |
| 2025-07-15 | moonshotai/Kimi-K2-Instruct | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-07-15). Open in app |
| 2025-07-10 | tomaarsen/Qwen3-Reranker-0.6B-seq-cls | template | the chat template was removed (2025-07-10). Open in app |
| 2025-07-09 | HuggingFaceTB/SmolLM3-3B | template unrendered | the chat template text changed; the renderer could not run one of the two versions, so whether rendered prompts changed is not known (2025-07-09). Open in app |
| 2025-07-08 | HuggingFaceTB/SmolLM3-3B | template rendering-changed default system prompt tool block | the chat template changed and renders differently (system+user, multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-07-08).default system prompt: With no system message, the text rendered before the user's message changed: "...o reach the conclusion. ## Tools ### XML Tools None ### Python Tools None user" became "...o reach the conclusion. user".tool block: A tool list passed to the template reached the prompt before and is dropped now. Open in app |
| 2025-07-08 | HuggingFaceTB/SmolLM3-3B | generation-defaults | generation defaults changed (do_sample unset to true, temperature unset to 0.6, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2025-07-08). Open in app |
| 2025-07-08 | HuggingFaceTB/SmolLM3-3B | tokenizer | config.json eos_token_id 128001 to 128012, but generation_config.json still declares 128012, which generate() reads first, so default generation stops on the same tokens; only code that reads config.eos_token_id directly sees the new value (2025-07-08). Open in app |
| 2025-07-03 | jinaai/jina-embeddings-v4 | derived parameters 2.80B to 3.09B (+288.2M) only because the reconstruction follows the class name; no size key changed; model_type declared as "qwen2_5_vl_text" (was unset) (2025-07-03). Open in app | |
| 2025-06-21 | vikhyatk/moondream2 | structural | declared dtype "float16" to "bfloat16" (2025-06-21). Open in app |
| 2025-06-12 | amd/DeepSeek-R1-MXFP4 | tokenizer | tokenizer model_max_length 512 to 16384; truncation defaults change (2025-06-12). Open in app |
| 2025-06-06 | tomaarsen/Qwen3-Reranker-0.6B-seq-cls | structural tokenizer | architectures ["Qwen3ForCausalLM"] to ["Qwen3ForSequenceClassification"]; a different class is instantiated from the same id; declared dtype "bfloat16" to "float32"; config.json pad_token_id unset to 151643; the tokenizer's pad token, which training pads with, did not change (2025-06-06). Open in app |
| 2025-06-05 | Qwen/Qwen3-Embedding-4B | structural | architectures ["Qwen3Model"] to ["Qwen3ForCausalLM"]; a different class is instantiated from the same id (2025-06-05). Open in app |
| 2025-06-05 | Qwen/Qwen3-Embedding-8B | structural | architectures ["Qwen3Model"] to ["Qwen3ForCausalLM"]; a different class is instantiated from the same id (2025-06-05). Open in app |
| 2025-06-04 | ai-sage/Giga-Embeddings-instruct | structural | context window 32768 to 4096 tokens (2025-06-04). Open in app |
| 2025-06-03 | Qwen/Qwen3-Embedding-0.6B | structural tokenizer generation-defaults stop tokens stop tokens | EOS stop list [151645,151643] to 151643; generation no longer stops on 151645 and runs past it; context window 40960 to 32768 tokens; generation_config.json pad_token_id 151643 to unset; generate() fills finished sequences in a batch with this id, while training pads with the tokenizer's pad token, which did not change (2025-06-03).stop tokens: generation_config no longer stops on any token that ends an answer in the template (<|im_end|> was dropped).stop tokens: generation_config no longer stops on any token that ends a tool call in the template (<|im_end|> was dropped). Open in app |
| 2025-05-29 | deepseek-ai/DeepSeek-R1-0528-Qwen3-8B | structural | rope_scaling rewritten in a different format ({"factor":4,"original_max_position_embeddings":32768,"rop... to {"attn_factor":0.8782488562869419,"factor":4,"original_ma...); a loader that only reads the old format rejects or ignores the new one (2025-05-29). Open in app |
| 2025-05-29 | ai-sage/Giga-Embeddings-instruct | structural | layer count 27 to 36; derived parameters 2.61B to 3.30B (+693.7M, +1.39 GB of bf16 weights) (2025-05-29). Open in app |
| 2025-05-29 | deepseek-ai/DeepSeek-R1-0528 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-29). Open in app |
| 2025-05-20 | Qwen/Qwen3-8B-Base | tokenizer eos vs template | EOS changed from "<|im_end|>" to "<|endoftext|>"; anything fine-tuned or served against the old value stops at a different token (2025-05-20).eos vs template: The template ends an answer with <|im_end|>, and the tokenizer's eos_token is now <|endoftext|> (it was <|im_end|>, which the template used). An engine that stops on the tokenizer's eos alone does not stop at the end of the turn. Open in app |
| 2025-05-20 | Qwen/Qwen3-4B-Base | tokenizer eos vs template | EOS changed from "<|im_end|>" to "<|endoftext|>"; anything fine-tuned or served against the old value stops at a different token (2025-05-20).eos vs template: The template ends an answer with <|im_end|>, and the tokenizer's eos_token is now <|endoftext|> (it was <|im_end|>, which the template used). An engine that stops on the tokenizer's eos alone does not stop at the end of the turn. Open in app |
| 2025-05-20 | Qwen/Qwen3-1.7B-Base | tokenizer eos vs template | EOS changed from "<|im_end|>" to "<|endoftext|>"; anything fine-tuned or served against the old value stops at a different token (2025-05-20).eos vs template: The template ends an answer with <|im_end|>, and the tokenizer's eos_token is now <|endoftext|> (it was <|im_end|>, which the template used). An engine that stops on the tokenizer's eos alone does not stop at the end of the turn. Open in app |
| 2025-05-20 | Qwen/Qwen3-0.6B-Base | tokenizer eos vs template | EOS changed from "<|im_end|>" to "<|endoftext|>"; anything fine-tuned or served against the old value stops at a different token (2025-05-20).eos vs template: The template ends an answer with <|im_end|>, and the tokenizer's eos_token is now <|endoftext|> (it was <|im_end|>, which the template used). An engine that stops on the tokenizer's eos alone does not stop at the end of the turn. Open in app |
| 2025-05-19 | Qwen/Qwen3-8B-FP8 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-8B-AWQ | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-8B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-4B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-32B-AWQ | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-32B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-30B-A3B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-235B-A22B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-14B-FP8 | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-14B-AWQ | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-14B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-1.7B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-19 | Qwen/Qwen3-0.6B | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-19). Open in app |
| 2025-05-16 | jinaai/jina-embeddings-v4 | structural | derived parameters 3.09B to 2.80B (-288.3M) only because the reconstruction follows the class name; no size key changed; architectures ["ColQwen25Duo"] to ["JinaEmbeddingsV4Model"]; a different class is instantiated from the same id (2025-05-16). Open in app |
| 2025-05-09 | unsloth/Qwen3-1.7B-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-05-09). Open in app |
| 2025-05-08 | Qwen/Qwen3-8B-AWQ | structural | declared dtype "bfloat16" to "float16" (2025-05-08). Open in app |
| 2025-05-06 | fixie-ai/ultravox-v0_5-llama-3_2-1b | tokenizer | tokenizer_class "PreTrainedTokenizerFast" to "PreTrainedTokenizer" (2025-05-06). Open in app |
| 2025-05-06 | DavidAU/Qwen3-8B-64k-Context-2X-Josiefied-Uncensored | structural | context window 40960 to 65536 tokens (2025-05-06). Open in app |
| 2025-05-06 | DavidAU/Qwen3-8B-64k-Context-2X-Josiefied-Uncensored | structural | rope_scaling added ({"factor":2,"original_max_position_embeddings":32768,"typ...); positions are now scaled (2025-05-06). Open in app |
| 2025-05-03 | Qwen/Qwen3-8B-AWQ | structural tokenizer template generation-defaults rendering-changed | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS declared as 151643 (was unset); context window 65536 to 40960 tokens (2025-05-03). Open in app |
| 2025-05-02 | Qwen/Qwen3-14B-GGUF | template rendering-changed | the chat template changed and renders differently (tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-05-02). Open in app |
| 2025-05-01 | Qwen/Qwen3-32B-AWQ | structural tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS declared as 151643 (was unset); context window 65536 to 40960 tokens (2025-05-01). Open in app |
| 2025-05-01 | Qwen/Qwen3-14B-AWQ | structural tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS declared as 151643 (was unset); context window 65536 to 40960 tokens (2025-05-01). Open in app |
| 2025-04-29 | unsloth/Qwen3-1.7B-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-04-29). Open in app |
| 2025-04-29 | unsloth/Qwen3-1.7B-GGUF | template cosmetic | the chat template text changed but rendered byte-identically on the three fixed conversations (system+user, multi-turn, tool call), all of which end on a user turn; a conversation ending on an assistant turn (training data, prefill) was not rendered (2025-04-29). Open in app |
| 2025-04-28 | Qwen/Qwen3-8B-Base | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-04-28).default system prompt: With no system message, the text rendered before the user's message removed: "system You are a helpful assistant. user" became "user". Open in app |
| 2025-04-28 | Qwen/Qwen3-8B-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS declared as 151643 (was unset); generation defaults changed (do_sample unset to true, temperature unset to 0.6, top_k unset to 20, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2025-04-28). Open in app |
| 2025-04-28 | Qwen/Qwen3-4B-Base | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-04-28).default system prompt: With no system message, the text rendered before the user's message removed: "system You are a helpful assistant. user" became "user". Open in app |
| 2025-04-28 | Qwen/Qwen3-1.7B-Base | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-04-28).default system prompt: With no system message, the text rendered before the user's message removed: "system You are a helpful assistant. user" became "user". Open in app |
| 2025-04-28 | Qwen/Qwen3-14B-FP8 | tokenizer generation-defaults | EOS stop list 151645 to [151645,151643]; generation now also stops on 151643, and still stops on every token it stopped on before; generation_config.json BOS declared as 151643 (was unset); generation defaults changed (do_sample unset to true, temperature unset to 0.6, top_k unset to 20, top_p unset to 0.95); a call that does not pass these explicitly now generates differently (2025-04-28). Open in app |
| 2025-04-28 | Qwen/Qwen3-0.6B-Base | template rendering-changed default system prompt | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-04-28).default system prompt: With no system message, the text rendered before the user's message removed: "system You are a helpful assistant. user" became "user". Open in app |
| 2025-04-28 | Qwen/Qwen3-235B-A22B | structural | MLP width 8192 to 12288; our reconstruction builds the same graph from both configs (same parameter count and structure) (2025-04-28). Open in app |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | structural | MLP width 8192 to 6144; our reconstruction builds the same graph from both configs (same parameter count and structure) (2025-04-28). Open in app |
| 2025-04-24 | ibm-granite/granite-3.3-2b-instruct | template rendering-changed | the chat template changed and renders differently (multi-turn, tool-call); prompts rendered with the old one no longer match what the model now expects, so cached prompts and eval sets need re-rendering (2025-04-24). Open in app |
| 2025-04-24 | unsloth/Llama-3.2-3B-Instruct | tokenizer | tokenizer_class "PreTrainedTokenizerFast" to "PreTrainedTokenizer" (2025-04-24). Open in app |
Showing the newest 300 of 741. Every row as JSON.
Licence and access
296 model card licence changes, 22 LICENSE file changes and 0 gating changes across 208 models.
| Date | Model | Kind | What changed |
|---|---|---|---|
| 2026-11-05 | ConwayResearch/Underdog-Saluki-27B-1.0 | licence | model card licence changed from "apache-2.0" to "cc-by-nc-4.0"; use under the old terms needs re-checking against the new ones (2026-11-05). |
| 2026-11-05 | ConwayResearch/Underdog-Saluki-27B-1.0 | licence-file | the LICENSE file's text changed (2026-11-05). |
| 2026-11-02 | XHToken/Spark-X2.6-4B-GGUF | licence-file | a LICENSE file was added (2026-11-02). |
| 2026-10-12 | melcheikh/gemma-4-31B-it-qat-NVFP4-Blackwell | licence | model card licence changed from "gemma" to {"license":"apache-2.0","license_link":"https://ai.google.dev/gemma/docs/gemma_4_license"}; use under the old terms needs re-checking against the new ones (2026-10-12). |
| 2026-10-08 | kindlingai/glm-5.3-exlr8-k2-k3-k4 | licence-file | the LICENSE file's text changed (2026-10-08). |
| 2026-10-08 | kindlingai/glm-5.3-exlr8-k2-k3-k4 | licence-file | the LICENSE file's text changed (2026-10-08). |
| 2026-09-01 | webAI-Official/TwIL-LM3 | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/webAI-Official/webAI-ColVec1-4b/blob/main/LICENSE.md","license_name":"webai-non-commercial-license-ver.-1.0"} to {"license":"other","license_link":"https://huggingface.co/webAI-Official/TwIL-LM3/blob/main/LICENSE.md","license_name":"webai-non-commercial-license-ver.-1.0"}; use under the old terms needs re-checking against the new ones (2026-09-01). |
| 2026-09-01 | farbodtavakkoli/OTel-2.0-LLM-31B-IT | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://ai.google.dev/gemma/docs/gemma_4_license"}; use under the old terms needs re-checking against the new ones (2026-09-01). |
| 2026-08-29 | nvidia/Kimi-K2-Thinking-NVFP4 | licence | model card licence changed from {"license":"other","license_name":"modified-mit"} to {"license":"other","license_name":"nvidia-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-08-29). |
| 2026-08-28 | unsloth/GLM-5.3-GGUF | licence | model card licence changed from "other" to {"license":"other","license_name":"glm-5.3"}; use under the old terms needs re-checking against the new ones (2026-08-28). |
| 2026-08-28 | zai-org/GLM-5.3 | licence | model card licence changed from "other" to {"license":"other","license_name":"glm-5.3"}; use under the old terms needs re-checking against the new ones (2026-08-28). |
| 2026-08-28 | zai-org/GLM-5.3 | licence-file | the LICENSE file's text changed (2026-08-28). |
| 2026-08-28 | superwhisper/s1-mini-GGUF | licence-file | the LICENSE file's text changed (2026-08-28). |
| 2026-08-27 | OBLITERATUS/Ornith-1.5-9B-OBLITERATED | licence | model card licence declared as "mit" (was undeclared) (2026-08-27). |
| 2026-08-26 | nvidia/llama-nemotron-rerank-1b-v2 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://openmdw.ai/license/1-1/","license_name":"openmdw-1.1"}; use under the old terms needs re-checking against the new ones (2026-08-26). |
| 2026-08-26 | nvidia/llama-nemotron-rerank-1b-v2 | licence-file | the LICENSE file's text changed (2026-08-26). |
| 2026-08-26 | ornith-ai/Ornith-1.5-9B-NVFP4 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-26). |
| 2026-08-19 | ornith-ai/Ornith-1.5-397B-NVFP4 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-397B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-397B-FP8 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-397B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-397B-GGUF | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-397B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-MLX | licence | model card licence {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"} removed; the card no longer states terms (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-MLX | licence | model card licence declared as {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"} (was undeclared) (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-NVFP4 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-FP8 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-GGUF | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B-GGUF | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B | licence | model card licence changed from {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.5-9B/blob/main/LICENSE"} to {"license":"mit","license_link":"https://huggingface.co/ornith-ai/Ornith-1.5-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.5-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-35B-A3B-MLX | licence | model card licence "mit" removed; the card no longer states terms (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B-MLX-8bit | licence | model card licence "mit" removed; the card no longer states terms (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B-MLX | licence | model card licence "mit" removed; the card no longer states terms (2026-08-19). |
| 2026-08-19 | ornith-ai/Ornith-1.5-9B-MLX-4bit | licence | model card licence "mit" removed; the card no longer states terms (2026-08-19). |
| 2026-08-17 | superwhisper/s1-mini-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"other","license_link":"LICENSE","license_name":"s1-mini-license"}; use under the old terms needs re-checking against the new ones (2026-08-17). |
| 2026-08-17 | superwhisper/s1-mini-GGUF | licence-file | the LICENSE file's text changed (2026-08-17). |
| 2026-08-15 | 0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF | licence | model card licence changed from "other" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-08-15). |
| 2026-08-08 | CMSManhattan/JiRackUltra_1b | licence | model card licence changed from "other" to "mit"; use under the old terms needs re-checking against the new ones (2026-08-08). |
| 2026-08-06 | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | licence | model card licence declared as {"license":"other","license_link":"https://openmdw.ai/license/1-1/","license_name":"openmdw-1.1"} (was undeclared) (2026-08-06). |
| 2026-07-24 | zeroentropy/zerank-2-reranker | licence | model card licence changed from "cc-by-nc-4.0" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-07-24). |
| 2026-07-23 | bartowski/Kwaipilot_KAT-Coder-V2.5-Dev-GGUF | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-07-23). |
| 2026-07-03 | k2-fsa/OmniVoice | licence | model card licence "apache-2.0" removed; the card no longer states terms (2026-07-03). |
| 2026-06-25 | ornith-ai/Ornith-1.0-397B-FP8 | licence | model card licence changed from {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B/blob/main/LICENSE"} to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B-FP8/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-25 | ornith-ai/Ornith-1.0-35B-GGUF | licence | model card licence changed from {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B/blob/main/LICENSE"} to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-25 | ornith-ai/Ornith-1.0-9B-GGUF | licence | model card licence changed from {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B/blob/main/LICENSE"} to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-25 | ornith-ai/Ornith-1.0-397B-FP8 | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-25 | ornith-ai/Ornith-1.0-35B-GGUF | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-25 | ornith-ai/Ornith-1.0-9B-GGUF | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-25). |
| 2026-06-21 | ornith-ai/Ornith-1.0-9B | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-21). |
| 2026-06-21 | ornith-ai/Ornith-1.0-35B | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-06-21). |
| 2026-06-19 | z-lab/Qwen3.6-35B-A3B-DFlash | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-06-19). |
| 2026-06-16 | amd/MiniMax-M3-MXFP4 | licence | model card licence declared as {"license":"other","license_link":"LICENSE","license_name":"minimax-community"} (was undeclared) (2026-06-16). |
| 2026-06-15 | tvall43/Qwen3.6-14B-A3B-FableVibes-GGUF | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-06-15). |
| 2026-06-15 | tvall43/Qwen3.6-14B-A3B-FableVibes-GGUF | licence | model card licence "apache-2.0" removed; the card no longer states terms (2026-06-15). |
| 2026-05-24 | openbmb/MiniCPM5-1B | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-05-24). |
| 2026-05-21 | biohub/ESMC-6B | licence | model card licence changed from "mit" to {"license":["mit","other"],"license_link":"https://github.com/Biohub/esm/blob/main/THIRD_PARTY_NOTICE.md"}; use under the old terms needs re-checking against the new ones (2026-05-21). |
| 2026-05-13 | nvidia/Kimi-K2.5-NVFP4 | licence | model card licence changed from {"license":"other","license_name":"modified-mit"} to {"license":"other","license_name":"nvidia-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-05-13). |
| 2026-05-11 | nvidia/Gemma-4-26B-A4B-NVFP4 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-software-and-model-evaluation-license","license_name":"nvidia-software-and-model-evaluation-license"} to {"license":"apache-2.0","license_link":"https://ai.google.dev/gemma/apache_2","license_name":"apache-license-2.0"}; use under the old terms needs re-checking against the new ones (2026-05-11). |
| 2026-05-10 | antirez/deepseek-v4-gguf | licence | model card licence declared as "mit" (was undeclared) (2026-05-10). |
| 2026-05-08 | antirez/deepseek-v4-gguf | licence | model card licence "mit" removed; the card no longer states terms (2026-05-08). |
| 2026-05-01 | nvidia/Gemma-4-26B-A4B-NVFP4 | licence | model card licence changed from "apache-2.0" to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-software-and-model-evaluation-license","license_name":"nvidia-software-and-model-evaluation-license"}; use under the old terms needs re-checking against the new ones (2026-05-01). |
| 2026-04-28 | nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 | licence | model card licence declared as {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-agreement/","license_name":"nvidia-open-model-agreement"} (was undeclared) (2026-04-28). |
| 2026-04-25 | sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP | licence | model card licence changed from "other" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-04-25). |
| 2026-04-17 | MiniMaxAI/MiniMax-M2.7 | licence | model card licence changed from {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2.7/blob/main/LICENSE","license_name":"modified-mit"} to {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2.7/blob/main/LICENSE","license_name":"other"}; use under the old terms needs re-checking against the new ones (2026-04-17). |
| 2026-04-15 | nvidia/MiniMax-M2.5-NVFP4 | licence | model card licence changed from {"license":"other","license_name":"modified-mit"} to {"license":"other","license_name":"nvidia-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-04-15). |
| 2026-04-15 | nvidia/Gemma-4-31B-IT-NVFP4 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://ai.google.dev/gemma/apache_2","license_name":"apache-license-2.0"}; use under the old terms needs re-checking against the new ones (2026-04-15). |
| 2026-04-14 | MiniMaxAI/MiniMax-M2.7 | licence-file | the LICENSE file's text changed (2026-04-14). |
| 2026-03-16 | nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 | licence | model card licence changed from {"license":"other","license_link":"LICENSE","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-16). |
| 2026-03-16 | nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 | licence | model card licence declared as {"license":"other","license_link":"LICENSE","license_name":"nvidia-open-model-license"} (was undeclared) (2026-03-16). |
| 2026-03-16 | nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 | licence | model card licence {"license":"other","license_link":"LICENSE","license_name":"nvidia-open-model-license"} removed; the card no longer states terms (2026-03-16). |
| 2026-03-15 | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-15). |
| 2026-03-15 | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-15). |
| 2026-03-15 | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-15). |
| 2026-03-14 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-14). |
| 2026-03-14 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-14). |
| 2026-03-14 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 | licence | model card licence changed from {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"} to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","license_name":"nvidia-nemotron-open-model-license"}; use under the old terms needs re-checking against the new ones (2026-03-14). |
| 2026-03-03 | sarvamai/sarvam-30b | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-03-03). |
| 2026-02-27 | EleutherAI/pythia-14m | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-02-27). |
| 2026-02-12 | MiniMaxAI/MiniMax-M2.5 | licence | model card licence changed from {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2.1/blob/main/LICENSE","license_name":"modified-mit"} to {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2.5/blob/main/LICENSE","license_name":"modified-mit"}; use under the old terms needs re-checking against the new ones (2026-02-12). |
| 2026-02-12 | MiniMaxAI/MiniMax-M2.5 | licence | model card licence changed from "mit" to {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2.1/blob/main/LICENSE","license_name":"modified-mit"}; use under the old terms needs re-checking against the new ones (2026-02-12). |
| 2026-02-08 | jinaai/jina-embeddings-v5-text-nano | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2026-02-08). |
| 2026-02-08 | jinaai/jina-embeddings-v5-text-small | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2026-02-08). |
| 2026-02-03 | Qwen/Qwen3-Coder-Next-FP8 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-Coder-Next/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-02-03). |
| 2026-02-03 | Qwen/Qwen3-Coder-Next | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-Coder-Next/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2026-02-03). |
| 2026-02-01 | nvidia/Kimi-K2.5-NVFP4 | licence | model card licence declared as {"license":"other","license_name":"modified-mit"} (was undeclared) (2026-02-01). |
| 2026-01-27 | deepseek-ai/DeepSeek-OCR-2 | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-01-27). |
| 2026-01-27 | deepseek-ai/DeepSeek-OCR-2 | licence | model card licence changed from "apache-2.0" to "mit"; use under the old terms needs re-checking against the new ones (2026-01-27). |
| 2026-01-27 | deepseek-ai/DeepSeek-OCR-2 | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2026-01-27). |
| 2026-01-08 | Qwen/Qwen3-VL-Embedding-2B | licence | model card licence declared as "apache-2.0" (was undeclared) (2026-01-08). |
| 2026-01-07 | Qwen/Qwen3-VL-Embedding-2B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2026-01-07). |
| 2026-01-04 | LiquidAI/LFM2.5-1.2B-Instruct-GGUF | licence | model card licence declared as {"license":"other","license_link":"LICENSE","license_name":"lfm1.0"} (was undeclared) (2026-01-04). |
| 2026-01-04 | LiquidAI/LFM2.5-1.2B-Instruct-GGUF | licence | model card licence {"license":"other","license_link":"LICENSE","license_name":"lfm1.0"} removed; the card no longer states terms (2026-01-04). |
| 2025-12-02 | nvidia/Kimi-K2-Thinking-NVFP4 | licence | model card licence changed from "mit" to {"license":"other","license_name":"modified-mit"}; use under the old terms needs re-checking against the new ones (2025-12-02). |
| 2025-12-01 | deepseek-ai/DeepSeek-V3.2 | licence | model card licence declared as "mit" (was undeclared) (2025-12-01). |
| 2025-11-13 | MiniMaxAI/MiniMax-M2 | licence | model card licence changed from "mit" to {"license":"other","license_link":"https://github.com/MiniMax-AI/MiniMax-M2/blob/main/LICENSE","license_name":"modified-mit"}; use under the old terms needs re-checking against the new ones (2025-11-13). |
| 2025-10-27 | MiniMaxAI/MiniMax-M2 | licence | model card licence changed from "apache-2.0" to "mit"; use under the old terms needs re-checking against the new ones (2025-10-27). |
| 2025-10-17 | nvidia/llama-embed-nemotron-8b | licence-file | the LICENSE file's text changed (2025-10-17). |
| 2025-10-16 | nvidia/llama-nemotron-embed-1b-v2 | licence | model card licence changed from "other" to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"}; use under the old terms needs re-checking against the new ones (2025-10-16). |
| 2025-10-16 | nvidia/llama-nemotron-embed-1b-v2 | licence | model card licence declared as "other" (was undeclared) (2025-10-16). |
| 2025-10-05 | openbmb/MiniCPM-o-2_6 | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-10-05). |
| 2025-09-23 | ai-sage/Giga-Embeddings-instruct | licence | model card licence declared as "mit" (was undeclared) (2025-09-23). |
| 2025-09-23 | ai-sage/Giga-Embeddings-instruct | licence | model card licence "mit" removed; the card no longer states terms (2025-09-23). |
| 2025-09-22 | PrimeIntellect/Qwen3-0.6B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-09-22). |
| 2025-09-15 | openbmb/MiniCPM-V-4_5 | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-09-15). |
| 2025-09-15 | openbmb/MiniCPM-V-4_5 | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-09-15). |
| 2025-09-15 | openbmb/MiniCPM-V-4_5 | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-09-15). |
| 2025-08-25 | openbmb/MiniCPM-V-4_5 | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-08-25). |
| 2025-08-19 | jinaai/jina-embeddings-v4 | licence | model card licence "cc-by-nc-4.0" removed; the card no longer states terms (2025-08-19). |
| 2025-08-18 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | licence | model card licence changed from "other" to {"license":"other","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","license_name":"nvidia-open-model-license"}; use under the old terms needs re-checking against the new ones (2025-08-18). |
| 2025-08-18 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | licence | model card licence declared as "other" (was undeclared) (2025-08-18). |
| 2025-08-13 | PrimeIntellect/Qwen3-0.6B | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2025-08-13). |
| 2025-08-12 | PrimeIntellect/Qwen3-0.6B | licence | model card licence declared as "mit" (was undeclared) (2025-08-12). |
| 2025-08-06 | Qwen/Qwen3-4B-Instruct-2507-FP8 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507-FP8/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-08-06). |
| 2025-08-05 | openai/gpt-oss-20b | licence-file | the LICENSE file's text changed (2025-08-05). |
| 2025-08-05 | openai/gpt-oss-120b | licence-file | the LICENSE file's text changed (2025-08-05). |
| 2025-08-05 | Qwen/Qwen3-4B-Thinking-2507 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-08-05). |
| 2025-08-05 | Qwen/Qwen3-4B-Instruct-2507 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-08-05). |
| 2025-07-31 | unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-31). |
| 2025-07-31 | Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-31). |
| 2025-07-31 | Qwen/Qwen3-Coder-30B-A3B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-31). |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507-FP8/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-29). |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | licence-file | the LICENSE file's text changed (2025-07-29). |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | licence-file | the LICENSE file's text changed (2025-07-29). |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-29). |
| 2025-07-29 | Qwen/Qwen3-30B-A3B-Instruct-2507 | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-29). |
| 2025-07-28 | Qwen/Qwen3-30B-A3B-Instruct-2507-FP8 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-28). |
| 2025-07-21 | Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507-FP8/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-07-21). |
| 2025-07-11 | moonshotai/Kimi-K2-Instruct | licence | model card licence changed from "mit" to {"license":"other","license_name":"modified-mit"}; use under the old terms needs re-checking against the new ones (2025-07-11). |
| 2025-07-11 | moonshotai/Kimi-K2-Instruct | licence | model card licence declared as "mit" (was undeclared) (2025-07-11). |
| 2025-07-11 | moonshotai/Kimi-K2-Instruct | licence | model card licence "mit" removed; the card no longer states terms (2025-07-11). |
| 2025-07-08 | HuggingFaceTB/SmolLM3-3B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-07-08). |
| 2025-06-24 | jinaai/jina-embeddings-v4 | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2025-06-24). |
| 2025-06-05 | Qwen/Qwen3-Embedding-0.6B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-06-05). |
| 2025-06-03 | Qwen/Qwen3-Embedding-0.6B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-06-03). |
| 2025-05-29 | ai-sage/Giga-Embeddings-instruct | licence | model card licence declared as "mit" (was undeclared) (2025-05-29). |
| 2025-05-29 | ai-sage/Giga-Embeddings-instruct | licence | model card licence "mit" removed; the card no longer states terms (2025-05-29). |
| 2025-05-08 | Qwen/Qwen3-0.6B-GGUF | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-1.7B-GGUF/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-0.6B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-08). |
| 2025-05-05 | Qwen/Qwen3-4B-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-05). |
| 2025-05-05 | Qwen/Qwen3-0.6B-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-1.7B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-05). |
| 2025-05-03 | Qwen/Qwen3-8B-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-8B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-03). |
| 2025-05-03 | Qwen/Qwen3-8B-AWQ | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-03). |
| 2025-05-01 | Qwen/Qwen3-14B-GGUF | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-14B-GGUF/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-01). |
| 2025-05-01 | Qwen/Qwen3-32B-AWQ | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-32B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-01). |
| 2025-05-01 | Qwen/Qwen3-14B-AWQ | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-14B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-05-01). |
| 2025-04-28 | Qwen/Qwen3-8B-Base | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B-FP8 | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-8B-FP8/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | licence | model card licence {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B/blob/main/LICENSE"} removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-1.7B | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-1.7B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-1.7B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-4B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-14B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-14B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-32B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-32B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-30B-A3B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-235B-A22B | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-32B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-235B-A22B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-14B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B-FP8 | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | unsloth/Qwen3-4B-GGUF | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen3-4B/blob/main/LICENSE"} (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-32B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-4B-Base | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-1.7B-Base | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-235B-A22B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B-Base | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-235B-A22B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-14B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-30B-A3B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-32B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-14B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-4B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-8B-Base | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-4B-Base | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-1.7B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-1.7B-Base | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B-Base | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-04-28). |
| 2025-04-28 | Qwen/Qwen3-0.6B | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-04-28). |
| 2025-04-25 | OpenGVLab/InternVL3-1B | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"}; use under the old terms needs re-checking against the new ones (2025-04-25). |
| 2025-04-25 | OpenGVLab/InternVL3-8B | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"}; use under the old terms needs re-checking against the new ones (2025-04-25). |
| 2025-02-28 | apple/OpenELM-1_1B-Instruct | licence | model card licence changed from {"license":"other","license_link":"LICENSE","license_name":"apple-sample-code-license"} to {"license":"apple-amlr","license_link":"LICENSE","license_name":"apple-sample-code-license"}; use under the old terms needs re-checking against the new ones (2025-02-28). |
| 2025-02-25 | GSAI-ML/LLaDA-8B-Instruct | licence | model card licence changed from "apache-2.0" to "mit"; use under the old terms needs re-checking against the new ones (2025-02-25). |
| 2025-02-24 | microsoft/Phi-4-multimodal-instruct | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Phi-4-multimodal-instruct/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2025-02-24). |
| 2025-02-24 | apple/OpenELM-1_1B-Instruct | licence-file | the LICENSE file's text changed (2025-02-24). |
| 2025-02-21 | google/siglip2-giant-opt-patch16-384 | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-02-21). |
| 2025-02-11 | nomic-ai/nomic-embed-text-v2-moe | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-02-11). |
| 2025-02-07 | fixie-ai/ultravox-v0_5-llama-3_2-1b | licence | model card licence declared as "mit" (was undeclared) (2025-02-07). |
| 2025-01-23 | deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | licence | model card licence declared as "mit" (was undeclared) (2025-01-23). |
| 2025-01-23 | deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | licence | model card licence declared as "mit" (was undeclared) (2025-01-23). |
| 2025-01-23 | deepseek-ai/DeepSeek-R1-Distill-Qwen-14B | licence | model card licence declared as "mit" (was undeclared) (2025-01-23). |
| 2025-01-23 | deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B | licence | model card licence declared as "mit" (was undeclared) (2025-01-23). |
| 2025-01-21 | deepseek-ai/DeepSeek-R1 | licence | model card licence declared as "mit" (was undeclared) (2025-01-21). |
| 2025-01-15 | nomic-ai/CodeRankEmbed | licence | model card licence declared as "mit" (was undeclared) (2025-01-15). |
| 2025-01-14 | openbmb/MiniCPM-o-2_6 | licence | model card licence "apache-2.0" removed; the card no longer states terms (2025-01-14). |
| 2025-01-14 | openbmb/MiniCPM-o-2_6 | licence | model card licence declared as "apache-2.0" (was undeclared) (2025-01-14). |
| 2024-12-20 | ai-sage/Giga-Embeddings-instruct | licence | model card licence declared as "mit" (was undeclared) (2024-12-20). |
| 2024-12-20 | ai-sage/Giga-Embeddings-instruct | licence | model card licence "mit" removed; the card no longer states terms (2024-12-20). |
| 2024-12-13 | ai-sage/Giga-Embeddings-instruct | licence | model card licence declared as "mit" (was undeclared) (2024-12-13). |
| 2024-12-12 | ai-sage/Giga-Embeddings-instruct | licence | model card licence "mit" removed; the card no longer states terms (2024-12-12). |
| 2024-12-11 | microsoft/phi-4 | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/microsoft/phi-4/resolve/main/LICENSE"} to {"license":"mit","license_link":"https://huggingface.co/microsoft/phi-4/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-12-11). |
| 2024-12-11 | microsoft/phi-4 | licence-file | the LICENSE file's text changed (2024-12-11). |
| 2024-11-12 | jinaai/jina-clip-v2 | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2024-11-12). |
| 2024-11-11 | Qwen/Qwen2.5-Coder-3B-Instruct | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/main/LICENSE","license_name":"qwen-research"} to {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct/blob/main/LICENSE","license_name":"qwen-research"}; use under the old terms needs re-checking against the new ones (2024-11-11). |
| 2024-10-31 | HuggingFaceTB/SmolLM2-135M | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-31). |
| 2024-10-31 | HuggingFaceTB/SmolLM2-135M-Instruct | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-31). |
| 2024-10-31 | HuggingFaceTB/SmolLM2-1.7B-Instruct | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-31). |
| 2024-10-31 | HuggingFaceTB/SmolLM2-360M-Instruct | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-31). |
| 2024-10-31 | HuggingFaceTB/SmolLM2-360M | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-31). |
| 2024-10-15 | hugging-quants/gemma-2-9b-it-AWQ-INT4 | licence | model card licence declared as "gemma" (was undeclared) (2024-10-15). |
| 2024-10-08 | jinaai/jina-clip-v2 | licence | model card licence "cc-by-nc-4.0" removed; the card no longer states terms (2024-10-08). |
| 2024-10-04 | openai/whisper-large-v3-turbo | licence | model card licence changed from "apache-2.0" to "mit"; use under the old terms needs re-checking against the new ones (2024-10-04). |
| 2024-10-01 | openai/whisper-large-v3-turbo | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-10-01). |
| 2024-09-25 | unsloth/Llama-3.2-1B-Instruct | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | unsloth/Llama-3.2-3B-Instruct | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | meta-llama/Llama-3.2-1B | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | meta-llama/Llama-3.2-3B | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | meta-llama/Llama-3.2-1B-Instruct | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | meta-llama/Llama-3.2-3B-Instruct | licence | model card licence declared as "llama3.2" (was undeclared) (2024-09-25). |
| 2024-09-25 | Maple728/TimeMoE-50M | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2024-09-25). |
| 2024-09-25 | utter-project/EuroLLM-1.7B-Instruct | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-09-25). |
| 2024-09-24 | utter-project/EuroLLM-1.7B-Instruct | licence | model card licence "apache-2.0" removed; the card no longer states terms (2024-09-24). |
| 2024-09-20 | meta-llama/Llama-3.2-1B | licence-file | the LICENSE.txt file's text changed (2024-09-20). |
| 2024-09-19 | bartowski/Qwen2.5-32B-Instruct-GGUF | licence | model card licence declared as {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/LICENSE"} (was undeclared) (2024-09-19). |
| 2024-09-19 | bartowski/Qwen2.5-32B-Instruct-GGUF | licence | model card licence "apache-2.0" removed; the card no longer states terms (2024-09-19). |
| 2024-09-18 | Qwen/Qwen2.5-Coder-7B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-Coder-7B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-Coder-7B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-Coder-1.5B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-Coder-1.5B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-7B | licence | model card licence changed from {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-32B/blob/main/LICENSE"} to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-7B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-Math-1.5B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-Math-1.5B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-0.5B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-0.5B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-1.5B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-1.5B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-7B | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-32B/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-7B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-32B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-14B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-14B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-1.5B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-0.5B-Instruct | licence | model card licence changed from "apache-2.0" to {"license":"apache-2.0","license_link":"https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct/blob/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-72B-Instruct | licence-file | the LICENSE file's text changed (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-72B-Instruct-AWQ | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"} to {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct-AWQ/blob/main/LICENSE","license_name":"qwen"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-3B-Instruct | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/main/LICENSE","license_name":"tongyi-qianwen-research"} to {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/main/LICENSE","license_name":"qwen-research"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-18 | Qwen/Qwen2.5-72B-Instruct | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"tongyi-qianwen"} to {"license":"other","license_link":"https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE","license_name":"qwen"}; use under the old terms needs re-checking against the new ones (2024-09-18). |
| 2024-09-17 | bartowski/Qwen2.5-32B-Instruct-GGUF | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-09-17). |
| 2024-09-14 | stepfun-ai/GOT-OCR2_0 | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-09-14). |
| 2024-08-20 | microsoft/Phi-3.5-vision-instruct | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Phi-3.5-vision-instruct/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-08-20). |
| 2024-08-20 | microsoft/Phi-3.5-mini-instruct | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Phi-3.5-mini-instruct/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-08-20). |
| 2024-08-20 | microsoft/Phi-3.5-MoE-instruct | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Phi-3.5-MoE-instruct/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-08-20). |
| 2024-08-15 | NousResearch/Hermes-3-Llama-3.1-8B | licence | model card licence declared as "llama3" (was undeclared) (2024-08-15). |
| 2024-07-30 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-30). |
| 2024-07-30 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence "llama3.1" removed; the card no longer states terms (2024-07-30). |
| 2024-07-28 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-28). |
| 2024-07-28 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence "llama3.1" removed; the card no longer states terms (2024-07-28). |
| 2024-07-26 | BAAI/bge-reranker-v2.5-gemma2-lightweight | licence | model card licence changed from "apache-2.0" to "gemma"; use under the old terms needs re-checking against the new ones (2024-07-26). |
| 2024-07-24 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-24). |
| 2024-07-24 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence "llama3.1" removed; the card no longer states terms (2024-07-24). |
| 2024-07-23 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-23). |
| 2024-07-23 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence "llama3.1" removed; the card no longer states terms (2024-07-23). |
| 2024-07-23 | unsloth/Meta-Llama-3.1-8B-Instruct | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-23). |
| 2024-07-23 | bartowski/Meta-Llama-3.1-8B-Instruct-GGUF | licence | model card licence declared as "llama3.1" (was undeclared) (2024-07-23). |
| 2024-07-23 | meta-llama/Llama-3.1-405B-FP8 | licence | model card licence changed from "other" to "llama3.1"; use under the old terms needs re-checking against the new ones (2024-07-23). |
| 2024-07-23 | meta-llama/Prompt-Guard-86M | licence | model card licence changed from "other" to "llama3.1"; use under the old terms needs re-checking against the new ones (2024-07-23). |
| 2024-07-23 | meta-llama/Llama-3.1-8B | licence | model card licence changed from "other" to "llama3.1"; use under the old terms needs re-checking against the new ones (2024-07-23). |
| 2024-07-23 | meta-llama/Llama-3.1-8B-Instruct | licence | model card licence changed from "other" to "llama3.1"; use under the old terms needs re-checking against the new ones (2024-07-23). |
| 2024-07-23 | meta-llama/Llama-3.1-405B-FP8 | licence | model card licence declared as "other" (was undeclared) (2024-07-23). |
| 2024-07-23 | meta-llama/Prompt-Guard-86M | licence | model card licence declared as "other" (was undeclared) (2024-07-23). |
| 2024-07-22 | meta-llama/Llama-3.1-8B | licence | model card licence declared as "other" (was undeclared) (2024-07-22). |
| 2024-07-18 | RedHatAI/DeepSeek-Coder-V2-Lite-Instruct-FP8 | licence | model card licence changed from {"license":"other","license_link":"LICENSE","license_name":"deepseek-license"} to {"license":"other","license_link":"https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL","license_name":"deepseek-license"}; use under the old terms needs re-checking against the new ones (2024-07-18). |
| 2024-07-18 | RedHatAI/DeepSeek-Coder-V2-Lite-Instruct-FP8 | licence | model card licence declared as {"license":"other","license_link":"LICENSE","license_name":"deepseek-license"} (was undeclared) (2024-07-18). |
| 2024-07-16 | google/gemma-2-2b-it | licence | model card licence declared as "gemma" (was undeclared) (2024-07-16). |
| 2024-07-01 | microsoft/Phi-3-mini-128k-instruct | licence-file | the LICENSE file's text changed (2024-07-01). |
| 2024-07-01 | microsoft/Phi-3-mini-4k-instruct | licence-file | the LICENSE file's text changed (2024-07-01). |
| 2024-07-01 | PekingU/rtdetr_r101vd_coco_o365 | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-07-01). |
| 2024-06-19 | jinaai/jina-reranker-v2-base-multilingual | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2024-06-19). |
| 2024-06-15 | microsoft/Florence-2-base-ft | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Florence-2-base-ft/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-06-15). |
| 2024-06-15 | microsoft/Florence-2-base | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Florence-2-base/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-06-15). |
| 2024-06-15 | microsoft/Florence-2-large | licence | model card licence changed from "mit" to {"license":"mit","license_link":"https://huggingface.co/microsoft/Florence-2-large/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-06-15). |
| 2024-06-07 | deepseek-ai/DeepSeek-V2-Lite-Chat | licence | model card licence declared as {"license":"other","license_link":"https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL","license_name":"deepseek"} (was undeclared) (2024-06-07). |
| 2024-06-07 | deepseek-ai/DeepSeek-V2-Lite | licence | model card licence declared as {"license":"other","license_link":"https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL","license_name":"deepseek"} (was undeclared) (2024-06-07). |
| 2024-05-22 | unsloth/mistral-7b-v0.3-bnb-4bit | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-05-22). |
| 2024-05-15 | deepseek-ai/DeepSeek-V2-Lite | licence | model card licence "mit" removed; the card no longer states terms (2024-05-15). |
| 2024-05-04 | MahmoudAshraf/mms-300m-1130-forced-aligner | licence | model card licence declared as "cc-by-nc-4.0" (was undeclared) (2024-05-04). |
| 2024-05-04 | MahmoudAshraf/mms-300m-1130-forced-aligner | licence | model card licence "cc-by-nc-4.0" removed; the card no longer states terms (2024-05-04). |
| 2024-04-20 | IlyaGusev/saiga_llama3_8b | licence | model card licence declared as {"license":"other","license_link":"https://llama.meta.com/llama3/license/","license_name":"llama3"} (was undeclared) (2024-04-20). |
| 2024-03-17 | BAAI/bge-reranker-v2-gemma | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-03-17). |
| 2024-03-17 | BAAI/bge-reranker-v2-m3 | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2024-03-17). |
| 2024-02-12 | nomic-ai/nomic-embed-text-v1.5 | licence | model card licence declared as "apache-2.0" (was undeclared) (2024-02-12). |
| 2024-02-06 | microsoft/phi-2 | licence-file | the LICENSE file's text changed (2024-02-06). |
| 2024-01-09 | Qwen/Qwen-72B | licence | model card licence declared as {"license":"other","license_link":"https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT","license_name":"tongyi-qianwen-license-agreement"} (was undeclared) (2024-01-09). |
| 2024-01-05 | microsoft/phi-2 | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE","license_name":"microsoft-research-license"} to {"license":"mit","license_link":"https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE"}; use under the old terms needs re-checking against the new ones (2024-01-05). |
| 2024-01-05 | microsoft/phi-2 | licence-file | the LICENSE file's text changed (2024-01-05). |
| 2023-12-31 | TinyLlama/TinyLlama-1.1B-Chat-v1.0 | licence | model card licence declared as "apache-2.0" (was undeclared) (2023-12-31). |
| 2023-12-13 | microsoft/phi-2 | licence | model card licence changed from {"license":"other","license_link":"https://huggingface.co/microsoft/phi-1_5/resolve/main/Research%20License.docx","license_name":"microsoft-research-license"} to {"license":"other","license_link":"https://huggingface.co/microsoft/phi-2/resolve/main/LICENSE","license_name":"microsoft-research-license"}; use under the old terms needs re-checking against the new ones (2023-12-13). |
| 2023-11-18 | Falconsai/nsfw_image_detection | licence | model card licence changed from "mit" to "apache-2.0"; use under the old terms needs re-checking against the new ones (2023-11-18). |
| 2023-11-18 | Falconsai/nsfw_image_detection | licence | model card licence changed from "apache-2.0" to "mit"; use under the old terms needs re-checking against the new ones (2023-11-18). |
| 2023-10-09 | jinaai/jina-embeddings-v2-small-en | licence | model card licence declared as "apache-2.0" (was undeclared) (2023-10-09). |
Lineage, as declared
Each tracked model that declares a base is checked against that base's card: a repo that calls itself a quantization of a base with no edit in its name while its own id or tags say abliterated, uncensored or heretic, and a derivative that declares a licence outside its base's family. These are statements about the card and tags, not about the weights or how the model behaves. 20 of 322 models with a declared base show one today.
| Model | Declared base | Finding |
|---|---|---|
| 0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF | Qwen/Qwen3.8-27B | edit-declared-as-quantization Declares itself a quantization of Qwen/Qwen3.8-27B, whose id names no edit, while its own id or tags say "Heretic". The model tree shows Qwen/Qwen3.8-27B as the direct parent with nothing in between. |
| Abiray/DeepSeek-V4.1-Flash-1017-Uncensored-GSQ-RCO-IQ2_XXS-GGUF | ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF | edit-declared-as-quantization Declares itself a quantization of ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF, whose id names no edit, while its own id or tags say "Uncensored". The model tree shows ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF as the direct parent with nothing in between. |
| JonathanColetti/Qwen3.8-27B-Uncensored-GGUF | Qwen/Qwen3.8-27B | edit-declared-as-quantization Declares itself a quantization of Qwen/Qwen3.8-27B, whose id names no edit, while its own id or tags say "Uncensored". The model tree shows Qwen/Qwen3.8-27B as the direct parent with nothing in between. |
| OBLITERATUS/Ornith-1.5-9B-OBLITERATED | ornith-ai/Ornith-1.5-9B | edit-declared-as-quantization Declares itself a quantization of ornith-ai/Ornith-1.5-9B, whose id names no edit, while its own id or tags say "abliterated". The model tree shows ornith-ai/Ornith-1.5-9B as the direct parent with nothing in between. |
| OBLITERATUS/Qwen3.8-27B-OBLITERATED | Qwen/Qwen3.8-27B | edit-declared-as-quantization Declares itself a quantization of Qwen/Qwen3.8-27B, whose id names no edit, while its own id or tags say "abliterated". The model tree shows Qwen/Qwen3.8-27B as the direct parent with nothing in between. |
| antirez/qwen3.8-flash-next-gguf | Qwen/Qwen3.8-Flash-Next | licence-family-mismatch The declared base Qwen/Qwen3.8-Flash-Next is under other (qwen-community-1.0), and this derivative declares other (qwen-community-license-1.0). |
| ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 | licence-family-mismatch The declared base nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 is under other (openmdw-1.1), and this derivative declares other. |
| huihui-ai/Huihui-DeepSeek-V4-Flash-0731-abliterated-GGUF | deepseek-ai/DeepSeek-V4-Flash-0731 | edit-declared-as-quantization Declares itself a quantization of deepseek-ai/DeepSeek-V4-Flash-0731, whose id names no edit, while its own id or tags say "abliterated". The model tree shows deepseek-ai/DeepSeek-V4-Flash-0731 as the direct parent with nothing in between. |
| huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated-MTP-GGUF | Qwen/Qwen3.6-35B-A3B | edit-declared-as-quantization Declares itself a quantization of Qwen/Qwen3.6-35B-A3B, whose id names no edit, while its own id or tags say "abliterated". The model tree shows Qwen/Qwen3.6-35B-A3B as the direct parent with nothing in between. |
| huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF | Qwen/Qwen3.8-27B | edit-declared-as-quantization Declares itself a quantization of Qwen/Qwen3.8-27B, whose id names no edit, while its own id or tags say "abliterated". The model tree shows Qwen/Qwen3.8-27B as the direct parent with nothing in between. |
| jimmy31chen/RVN-Qwen3.8-Flash-Next-Abliterated-Uncensored-IQ3_S-Strata | 0bserverx/RVN-Qwen3.8-Flash-Next-Abliterated-Uncensored | licence-family-mismatch The declared base 0bserverx/RVN-Qwen3.8-Flash-Next-Abliterated-Uncensored is under other (qwen-community-1.0), and this derivative declares apache-2.0. |
| nvidia/GLM-5.3-NVFP4 | zai-org/GLM-5.3 | licence-family-mismatch The declared base zai-org/GLM-5.3 is under other (glm-5.3), and this derivative declares other (nvidia-open-model-license). |
| nvidia/Kimi-K2-Thinking-NVFP4 | moonshotai/Kimi-K2-Thinking | licence-family-mismatch The declared base moonshotai/Kimi-K2-Thinking is under other (modified-mit), and this derivative declares other (nvidia-open-model-license). |
| nvidia/Kimi-K2.5-NVFP4 | moonshotai/Kimi-K2.5 | licence-family-mismatch The declared base moonshotai/Kimi-K2.5 is under other (modified-mit), and this derivative declares other (nvidia-open-model-license). |
| nvidia/LocateAnything-3B | Qwen/Qwen2.5-3B-Instruct | licence-family-mismatch The declared base Qwen/Qwen2.5-3B-Instruct is under other (qwen-research), and this derivative declares other (nvidia-license). |
| nvidia/MiniMax-M2.5-NVFP4 | MiniMaxAI/MiniMax-M2.5 | licence-family-mismatch The declared base MiniMaxAI/MiniMax-M2.5 is under other (modified-mit), and this derivative declares other (nvidia-open-model-license). |
| nvidia/MiniMax-M3-NVFP4 | MiniMaxAI/MiniMax-M3 | licence-family-mismatch The declared base MiniMaxAI/MiniMax-M3 is under other (minimax-community), and this derivative declares other (minimax-community-license). |
| nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 | nvidia/NVIDIA-Nemotron-Nano-9B-v2 | licence-family-mismatch The declared base nvidia/NVIDIA-Nemotron-Nano-9B-v2 is under other (nvidia-open-model-license), and this derivative declares other (nvidia-nemotron-open-model-license). |
| nvidia/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-NVFP4 | nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | licence-family-mismatch The declared base nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 is under other (nvidia-nemotron-open-model-license), and this derivative declares other (openmdw-1.1). |
| soyaakinohara/Spark-X2.5-4B-Heretic-jp-gguf | XHToken/Spark-X2.5-4B | edit-declared-as-quantization Declares itself a quantization of XHToken/Spark-X2.5-4B, whose id names no edit, while its own id or tags say "Heretic". The model tree shows XHToken/Spark-X2.5-4B as the direct parent with nothing in between. |
How this is read
Every day the tracker asks the Hugging Face Hub for the head commit of each tracked model. Only when that commit moved does it read config.json, generation_config.json, tokenizer_config.json, chat_template.jinja and special_tokens_map.json at the new commit, keep a normalised subset of the keys that change what the model computes, how it tokenises, or how it generates by default, and diff it against the last snapshot.
History before the first daily run comes from walking each repository's git history for those files, which is complete for them and blind to the weights: a weight swap with no file change is only visible from the first daily snapshot on.
Chat templates are stored as a sha256 prefix, never in full. Aliases are folded (torch_dtype and dtype, rope_theta inside rope_parameters) so a re-save under a newer transformers is not reported as a change.
A key that differs as stored but not in meaning is kept and labelled format-only, with the fold that equates the two values: RoPE settings rewritten in the transformers v5 rope_parameters form, a tokenizer class and its Fast alias, the library's "unset" sentinel for model_max_length, stop-token ids written as a scalar or a list, and special tokens moved between special_tokens_map.json and tokenizer_config.json. A row whose every key is format-only is a re-save, and it does not count a model as changed.
When a chat template changes, both versions are rendered with @huggingface/jinja on three fixed conversations (system and user, multi-turn, a tool call). Cosmetic means all three rendered byte-identically; that is a statement about those inputs, not a proof the templates are equivalent on every input. The comparison is of rendered text, not token ids: with an unchanged tokenizer the two agree.
Safety-relevant config changes are labelled on the row they belong to: the default system prompt, the tool block, the generation prefix, role validation, a template branch on a non-role literal or a new URL, template code outside the safe Jinja subset, the tokenizer eos_token no longer ending a turn the template writes, and the generation stop list starting or stopping to cover such a token. Each label carries its basis. They are statements about the configuration files and what the renderer printed, never a judgement of how the model behaves; a row without a label was either not read for them or showed nothing, which is not a clean bill.
Licence and access are their own lane: the model card licence field (history recovered from the README's git history), the top-level LICENSE file, and the Hub's gated flag. Gating has no git history, so gating changes exist only from the first daily read on.
Parameter and memory deltas come from reconstructing the graph from each config with the same code the /m/ pages use. Every consequence sentence says what the key does mechanically; none is a report that something broke for anyone.
Pin what you tested
from_pretrained("XiaomiMiMo/MiMo-V2.5", revision="d6bf9db68673d1f903a53d8da417b293fdc97d1f")