Conversation
remove inference-server submodule and route benchmarking to artifact repo inetgrate inference server as artifact
Contributor
There was a problem hiding this comment.
Pull request overview
This PR removes the tt-inference-server git submodule and switches TT Studio to consume tt-inference-server as a versioned release artifact, while introducing a dedicated inference-api FastAPI service that wraps the artifact’s run.py.
Changes:
- Remove the
tt-inference-serversubmodule and.gitmodulesentry, replacing it with an artifact-based workflow managed from.artifacts/tt-inference-server. - Update
run.pyto download and extract a configurablett-inference-serverrelease tarball, wire environment variables for the artifact, and start aninference-apiFastAPI app instead of the old submodule-based server. - Add a new
inference-apipackage (FastAPI app, requirements, API endpoints) that integrates with the artifact’srun.pyand workflows, including deployment progress and log streaming endpoints.
Reviewed changes
Copilot reviewed 8 out of 9 changed files in this pull request and generated 7 comments.
Show a summary per file
| File | Description |
|---|---|
| tt-inference-server | Removes the git submodule reference so the repo no longer embeds tt-inference-server directly. |
| run.py | Switches setup from git submodules to downloaded artifacts, configures the artifact directory/ENV, and starts the new inference-api instead of the embedded FastAPI server. |
| inference-api/requirements.txt | Defines dependencies for the new inference API (FastAPI, uvicorn, docker, pydantic, etc.). |
| inference-api/main.py | Entry point exposing the FastAPI app from api.py for uvicorn. |
| inference-api/api.py | New FastAPI service that imports run.py and workflows from the artifact, manages deployment runs, progress/log tracking, and exposes model/workflow/device listing endpoints. |
| inference-api/init.py | Package marker file with SPDX headers. |
| app/.env.default | Adds TT_INFERENCE_ARTIFACT_VERSION default to configure which tt-inference-server release to use. |
| .gitmodules | Deletes the tt-inference-server submodule configuration. |
| .gitignore | Ignores .artifacts, common archive formats, and inference-api virtualenv/bytecode. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
anirudTT
reviewed
Feb 6, 2026
| "last_updated": time.time() | ||
| }) | ||
|
|
||
| # Add stalled detection (>120s no updates) |
Collaborator
There was a problem hiding this comment.
I think we need to change this logic , i think this might be causing deployment issues for larger models .
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
…torrent/tt-studio into rumeza/externalize-tt-infer
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Collaborator
|
* Enhance inference server configuration with reconfiguration options and improved artifact info logging. Added support for reconfiguring inference server artifacts and updated metadata writing to include detailed configuration information and instructions. * Implement artifact re-download logic based on version and branch changes in inference server configuration. Added proactive sudo authentication request for artifact management to prevent permission issues during cleanup.
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
2 participants
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.
This PR pivots the usage of tt-inference-server to a release artifact instead
How the artifact is created:
The artifact is a .tar.gz of the tt-inference-server repository source code.
GitHub releases: When a release/tag is created in the tt-inference-server repo, GitHub automatically generates a source code tarball at: https://github.com/tenstorrent/tt-inference-server/archive/refs/tags/{version}.tar.gz
the run.py downloads this tarball and extracts it to .artifacts/tt-inference-server/.
TT_INFERENCE_ARTIFACT_VERSION=v0.8.0
you can change up the version of tt-inference by changing in env
