[Misc] Delete LoRA-related redundancy code#17841
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
|
Could you explain in the PR description how is it redundant? |
| "embed_tokens", | ||
| "lm_head", | ||
| ] | ||
| embedding_modules = { |
There was a problem hiding this comment.
Static var supported_lora_modules has been deprecated for some time, so delete it directly.
| org_num_embeddings=config.vocab_size, | ||
| padding_size=DEFAULT_VOCAB_PADDING_SIZE | ||
| # We need bigger padding if using lora for kernel compatibility | ||
| if not lora_config else lora_config.lora_vocab_padding_size, |
There was a problem hiding this comment.
Grok's LoRA probably won't have scenarios involving expanded vocabulary, so this code is also useless.
| DEFAULT_VOCAB_PADDING_SIZE | ||
| # We need bigger padding if using lora for kernel | ||
| # compatibility | ||
| if not lora_config else lora_config.lora_vocab_padding_size), |
DarkLight1337
left a comment
There was a problem hiding this comment.
LGTM, thanks for the explanation!
|
Retrying LoRA TP test. If it still doesn't pass then can you fix it? |
These changes should not affect the LoRA test. |
|
Ok the TP test passed, it would be great if you could update the test to avoid flakiness in another PR! |
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Yuqi Zhang <yuqizhang@google.com>
No description provided.