Update test of Youtu-LLM to pr-aligned repos#43578
Conversation
|
[For maintainers] Suggested jobs to run (before merge) run-slow: youtu |
molbap
left a comment
molbap
left a comment
There was a problem hiding this comment.
seems good - let's rerun slow tests
| tokenizer = AutoTokenizer.from_pretrained("tencent/Youtu-LLM-2B-Base") | ||
| model = YoutuForCausalLM.from_pretrained( | ||
| "Junrulu/Youtu-LLM-2B-Base-hf", device_map=torch_device, dtype=torch.float16 | ||
| "tencent/Youtu-LLM-2B-Base", device_map=torch_device, dtype=torch.float16 |
There was a problem hiding this comment.
noticed the dtype on the config seems to be bfloat16, is this dtype intended here?
There was a problem hiding this comment.
dtype=float16 is used to follow the paradigm of other existing tests (e.g., Qwen3, DeepseekV3), as we hope to keep consistent with them.
p.s. These models, as well as Youtu-LLM, are all set dtype as bfloat16 in the config (Qwen3 config, DeepseekV3 config).
There was a problem hiding this comment.
hmm I see. It's likely not a big deal but I wonder why we set them up like this, in particular for inference. Thanks for answering!
|
run-slow: youtu |
|
This comment contains models: ["models/youtu"] |
CI Results✅ No failing test specific to this PR 🎉 ! |
2 participants
What does this PR do?
To update the test of Youtu-LLM to pr-aligned repos, fixing issues mentioned here: #43166 (comment).
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@molbap @vasqu @xenova