Skip to content

E2e model quality upstream - #853

Draft
yannicks1 wants to merge 24 commits into
torch-spyre:mainfrom
yannicks1:e2e-model-quality-upstream
Draft

E2e model quality upstream#853
yannicks1 wants to merge 24 commits into
torch-spyre:mainfrom
yannicks1:e2e-model-quality-upstream

Conversation

@yannicks1

Copy link
Copy Markdown
Collaborator

Description

Related Issues

Test Plan

Checklist

  • I have read the contributing guidelines
  • My code follows the project's code style (run bash format.sh)
  • I have added tests for my changes (if applicable)
  • I have updated the documentation (if applicable)
  • My commits include a Signed-off-by: line (DCO compliance)

Compare compiled Spyre output for the product decoder and embedding models
against cached CPU HF references, under a new model_quality marker with its
own CI job. gemma-4-31B is gated on short prompts only: it drifts from HF as
the prompt grows because torch-spyre normalises in fp16.

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
…utions

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
…ill bucket

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
…t step coverage

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
…sion as unpinned

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
… refs

Drop the checked-in reference JSONs and the scripts that generated them. Each
gate now computes its HF side in-run through upstream vLLM's HfRunner and judges
it with upstream's own check_logprobs_close / check_embeddings_close, reached
through the pinned tests/ tree the upstream suite already clones.

That clone helper moves out of pytest_plugin.py into spyre_testing_plugin/
upstream.py. Since tests.* now resolves to upstream's package rather than this
repo's directory, sibling test modules import each other as top-level modules.

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
@github-actions

Copy link
Copy Markdown
Contributor

👋 Hi! Thank you for contributing.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, run ./format.sh.
Now you are good to go 🚀.

We also recommend installing prek and configuring it to check your code before every local commit.

@github-actions

github-actions Bot commented Sep 10, 2026

Copy link
Copy Markdown
Contributor

📖 This PR includes doc changes. Download the preview (built from 9b616e5), or build locally:

uv pip install -r docs/requirements-docs.txt
mkdocs serve

Signed-off-by: Yannick Schnider <Yannick.Schnider1@ibm.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant