Changes / Feeds / IlyaGusev/saiga_llama3_8b
IlyaGusev/saiga_llama3_8b
4 changes to a tracked key since 2024-04-18, newest first. Head as last read: 5bb9917bdb85340549662ebb62c8e522037ff3f3.
| Date | Advisory | What changed, and what it does |
|---|---|---|
| 2024-04-26 | x_NXMC-2024-0172 breaking | EOS changed from "<|im_end|>" to "<|eot_id|>"; anything fine-tuned or served against the old value stops at a different token; BOS changed from "<|im_start|>" to "<|begin_of_text|>"; prompts built by the old tokenizer start with a different token; pad token "<|im_start|>" to "<|begin_of_text|>"; batched training masks a different id (2024-04-26).91d7b8a51d0d -> bc54c81cc63a |
| 2024-04-25 | x_NXMC-2024-0171 breaking | EOS changed from 128001 to 128009; anything fine-tuned or served against the old value stops at a different token; pad token "<|reserved_special_token_0|>" to "<|im_start|>"; batched training masks a different id (2024-04-25).d04705b3c527 -> 91d7b8a51d0d |
| 2024-04-19 | x_NXMC-2024-0170 breaking | 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 (2024-04-19).0af818c24e23 -> d04705b3c527 |
| 2024-04-19 | x_NXMC-2024-0169 breaking | pad token unset to 128002; batched training masks a different id; generation defaults changed (do_sample unset to true, max_new_tokens unset to 1536, repetition_penalty unset to 1.12, temperature unset to 0.2); a call that does not pass these explicitly now generates differently (2024-04-19).f4eb6c87c589 -> 0af818c24e23 |
Pin it, and let a bot move the pin
from_pretrained("IlyaGusev/saiga_llama3_8b", revision="5bb9917bdb85340549662ebb62c8e522037ff3f3")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/IlyaGusev/saiga_llama3_8b.html)