Changes / Feeds / Qwen/Qwen3-235B-A22B-Instruct-2507-FP8
Qwen/Qwen3-235B-A22B-Instruct-2507-FP8
4 changes to a tracked key since 2025-07-21, newest first. Head as last read: e156cb4efae43fbee1a1ab073f946a1377e6b969.
| Date | Advisory | What changed, and what it does |
|---|---|---|
| 2025-09-17 | x_NXMC-2025-0155 | 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).652b9036a3d9 -> e156cb4efae4 Open in app |
| 2025-07-21 | x_NXMC-2025-0154 breaking | 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).c581b8ffb8fd -> 652b9036a3d9 Open in app |
| 2025-07-21 | x_NXMC-2025-0153 | 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).0818ef1ddff8 -> c581b8ffb8fd Open in app |
| 2025-07-21 | x_NXMC-2025-0152 breaking | 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).d29dc93a6c15 -> b939f17ff56d Open in app |
Pin it, and let a bot move the pin
from_pretrained("Qwen/Qwen3-235B-A22B-Instruct-2507-FP8", revision="e156cb4efae43fbee1a1ab073f946a1377e6b969")
With the Renovate preset, a pull request moves this pin when the head moves, and links back here.
[](https://www.neurarch.com/changes-feeds/models/Qwen/Qwen3-235B-A22B-Instruct-2507-FP8.html)