[Model] Optimize nemotron_h implementation #19249
Conversation
Signed-off-by: Jee Jee Li <[email protected]>
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
| "k_proj", | ||
| "v_proj", | ||
| ], | ||
| "gate_up_proj": ["up_proj", "down_proj"] |
There was a problem hiding this comment.
It's incorrect property, delete it
| ) -> None: | ||
| super().__init__() | ||
| self.up_proj = MergedColumnParallelLinear( | ||
| self.up_proj = ColumnParallelLinear( |
There was a problem hiding this comment.
Use ColumnParallelLinear, there's no need to use MergedColumnParallelLinear
|
👋 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 🚀 |
github-actions
bot
added
the
ready
2 participants
Essential Elements of an Effective PR Description Checklist
Purpose
Test Plan
Test Result