vllm-project/tpu-inference

[Enhancement]: Add unit test for loading Deepseek-V3 model

Open

#1,852 opened on Mar 4, 2026

View on GitHub
 (3 comments) (0 reactions) (0 assignees)Python (205 forks)github user discovery
contribution-welcomeenhancementgood first issuehelp wanted

Repository metrics

Stars
 (348 stars)
PR merge metrics
 (PR metrics pending)

Description

🚀 The feature, motivation and pitch

Qwen models have test_model_loading to verify the model can successfully load. Deepseek_v3 is lacking such unit test.

I created https://huggingface.co/hugg1ngfac3/deepseek-r1-tiny which is a trimmed version of DS-v3, for testing.

This request is to add unit test (test_model_loading specifically) for Deepseek-v3:

  • start with non-MLA, see necessary flags in #1835
  • follow-up with MLA, after #1845 merged
  • start with sanity check, correctness check is bonus

Alternatives

No response

Additional context

No response

Before submitting a new issue...

  • Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

Contributor guide

Research direction
Look at the existing test for Qwen MoE model loading in tests/models/jax/test qwen3 moe.py and create a similar test for Deepseek v3 using the provided tiny model from HuggingFace. Start with non MLA configuration, then follow up with MLA after PR #1845 is merged.
Tech stack
python
Domain
backendtesting qa
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Fresh
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonUnit testing
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
75