[GH-ISSUE #6060] Update template: Llama 3.1 #50299

Closed
opened 2026-04-28 15:00:14 -05:00 by GiteaMirror · 4 comments
Owner

Originally created by @MaxJa4 on GitHub (Jul 29, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6060

Meta / HF updated the tokenizer config (specifically the chat template) of all the Llama 3.1 (instruct) models a few hours ago:

Since there have been some reports of quality issues with the models themselves (not with Ollama specifically), it may help fix things.
The new chat templates are quite different from the original ones.
Just as a heads up, so it doesn't get overlooked.

Thanks! :)

Update: it changed again

Originally created by @MaxJa4 on GitHub (Jul 29, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6060 Meta / HF updated the tokenizer config (specifically the chat template) of all the Llama 3.1 (instruct) models a few hours ago: - [Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct/commit/b2a4d0f33b41fcd59a6d31662cc63b8d53367e1e) - [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct/commit/168b5b277b07436c229bd69844a73872eb8b73a8) - [Meta-Llama-3.1-405B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-Instruct/commit/f9801cba95a53242b3cc928a4a418d12571d1c5f) Since there have been some reports of quality issues with the models themselves (not with Ollama specifically), it may help fix things. The new chat templates are quite different from the original ones. Just as a heads up, so it doesn't get overlooked. Thanks! :) Update: it changed again
GiteaMirror added the model label 2026-04-28 15:00:14 -05:00
Author
Owner

@d-kleine commented on GitHub (Jul 30, 2024):

When updating, please also consider adding #5957

<!-- gh-comment-id:2258728207 --> @d-kleine commented on GitHub (Jul 30, 2024): When updating, please also consider adding #5957
Author
Owner

@igorschlum commented on GitHub (Aug 6, 2024):

Llama3.1 are being updated since yesterday: https://ollama.com/library/llama3.1/tags

<!-- gh-comment-id:2270604607 --> @igorschlum commented on GitHub (Aug 6, 2024): Llama3.1 are being updated since yesterday: https://ollama.com/library/llama3.1/tags
Author
Owner

@d-kleine commented on GitHub (Aug 6, 2024):

@igorschlum Would it be possible to add the 405b text completion LLama 3.1 model there as well? Curently there is only the instruct fine-tuned version of the 405b model.

Edit: Now there, thanks!

<!-- gh-comment-id:2271938799 --> @d-kleine commented on GitHub (Aug 6, 2024): @igorschlum Would it be possible to add the 405b `text` completion LLama 3.1 model there as well? Curently there is only the `instruct` fine-tuned version of the 405b model. Edit: Now there, thanks!
Author
Owner

@MaxJa4 commented on GitHub (Aug 7, 2024):

Seems like everything is done here, thanks!

<!-- gh-comment-id:2273861728 --> @MaxJa4 commented on GitHub (Aug 7, 2024): Seems like everything is done here, thanks!
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#50299