Add GLM-5.2 low-latency serving recipe - #652
Draft
simon-mo wants to merge 1 commit into
Draft
Conversation
Signed-off-by: simon-mo <simon.mo@hey.com>
Contributor
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Contributor
There was a problem hiding this comment.
Code Review
This pull request updates the GLM-5.2 model recipe to support the low-latency NVFP4 variant on Blackwell GPUs. It introduces specific installation notes, prerequisites, and detailed Docker/CLI command configurations for NVFP4, including MTP speculative decoding, model class overrides, and breakable piecewise CUDA graphs. Additionally, the recipe build script (build-recipes-api.mjs) is updated to automatically enable features forced by a variant's default_modes. There are no review comments, and I have no feedback to provide.
|
@simon-mo Fix the merge conflict |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
vllm/vllm-openai:v0.25.1.:glm52guide example with the release-tested copy-paste command.default_modeswhen building the static JSON API.Why
The public GLM-5.2 guide still showed the older image and a generic NVFP4 command, while the current low-latency path requires vLLM 0.25.1 and explicit fused-model/CUDA-graph configuration.
The API generator also excluded opt-in features unconditionally. As a result, promoted variant endpoints dropped MTP even when a variant's
default_modesexplicitly enabled it. The generator now mirrors the interactive command builder.Impact
Selecting the NVFP4 variant on B200/B300, or consuming
/nvidia/GLM-5.2-NVFP4.json, produces the same low-latency command with MTP=5 and the pinned Docker image.Validation
node scripts/build-recipes-api.mjspublic/nvidia/GLM-5.2-NVFP4.json.git diff --checkpnpm lintwas not available as a non-interactive check because this repository has no ESLint configuration andnext lintopens the first-run configuration prompt.