Add integration tests for PyTorch, TGI and TEI DLCs #79
Conversation
This file contains 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
As it will be reused within the TGI and TEI tests
Pass args via `text_generation_launcher_kwargs` and include the VertexAI environment mimic via the `AIP_` environment variables.
alvarobartt
force-pushed
the
add-integration-tests
branch
from
157ab15 to
9446a3e
Compare
Which is odd, since `jinja2` is a core dependency of `transformers`, see https://github.com/huggingface/transformers/blob/174890280b340b89c5bfa092f6b4fb0e2dc2d7fc/setup.py#L127
alvarobartt
changed the title
philschmid
reviewed
Sep 2, 2024
Comment on lines
37
to
39
| training-dlc: us-docker.pkg.dev/deeplearning-platform-release/gcr.io/huggingface-pytorch-training-cu121.transformers.4-42.ubuntu2204.py310 | ||
| inference-dlc: us-docker.pkg.dev/deeplearning-platform-release/gcr.io/huggingface-pytorch-inference-cu121.2-2.transformers.4-44.ubuntu2204.py311 | ||
| tgi-dlc: us-docker.pkg.dev/deeplearning-platform-release/gcr.io/huggingface-text-generation-inference-cu121.2-2.ubuntu2204.py310 |
There was a problem hiding this comment.
Mhm is there a better way to specify those? Feels like we can easily forget updating them?
- Capture `container_uri` from environment variable before running testand remove the default value to prevent issues when testing - Remove `max_train_epochs=-1` as not required since `max_steps` isalready specified - Rename `test_transformers` to `test_huggingface_inference_toolkit` - Remove `transformers` and `jinja2` dependencies as not required, as well as `AutoTokenizer` usage for prompt formatting Co-authored-by: Philipp Schmid <[email protected]>
alvarobartt
force-pushed
the
add-integration-tests
branch
from
3af2bcf to
7ce5aeb
Compare
alvarobartt
force-pushed
the
add-integration-tests
branch
from
6eb06b5 to
349df29
Compare
…ia-smi` Those dependencies where not needed, not actively maintained and adding extra complexity; instead, it has been replaced with `subprocess` running `nvidia-smi`.
- TEI condition on container port was reversed - `gpu_available` raises exception instead of `returncode` if command doesn't exist
In most of the cases, splitting those is for the best and to reduce execution time, assuming we tend to update the DLCs one at a time, so it's not really probable for all the containers to change at once. Pros: easier to manage, more granular, no need for extra `docker pull`s, just runs what's modified Cons: when modifying a bunch of tests it will be slower as `docker pull` needs to be done per each test as instances are ephemeral
The `type: choice` with `options` is only supported for `workflow_dispatch` i.e. when triggering the GitHub Action manually; not via `workflow_call` i.e. when the workflow is just reused from another workflow.
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.
Description
This PR adds some integration tests for the following Hugging Face DLCs on Google Cloud:
The tests related to the inference try different alternatives, as well as emulate the Vertex AI environments via the
AIP_environment variables exposed by Vertex AI and handled within the Hugging Face DLCs on Google Cloud for a seamless integration.