[Llama4 Guard] Add JAX Llama-Guard-4-12B Text Portion #1090
Conversation
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
…output of the tokenizer.encode() call in the inference script
…. No longer have any changes in vllm
…_template, removing the need for the .jinja file
…tances from Llama Guard 4 subclass initializations to comply with unit tests
…ertion error from model downsizing
…into jiries/llama-guard-4-text
…ile and replaced the model_loader registry key name for Llama Guard 4 with a recognized text only model
…into jiries/llama-guard-4-text
…into jiries/llama-guard-4-text
…into jiries/llama-guard-4-text
DescriptionStart with a short description of what the PR does and how this is a change from The rest of the description includes relevant details and context, examples:
If the change fixes a bug or a Github issue, please include a link, e.g.,: TestsPlease describe how you tested this change, and include any instructions and/or ChecklistBefore submitting this PR, please make sure:
|
jrplatin
reviewed
Nov 13, 2025
Collaborator
|
Please update the title + add accuracy / performance testing in the description too |
JiriesKaileh
changed the title
jrplatin
changed the title
jrplatin
changed the title
jrplatin
reviewed
Nov 14, 2025
…atement in TPUModelRunner.load_model()
…verride in CI scripts, and simplified prompt formatting in offline inference script
…nference script, and made minor change to CI scripts to prevent breaking CI
…into jiries/llama-guard-4-text
jrplatin
reviewed
Nov 17, 2025
jrplatin
reviewed
Nov 17, 2025
jrplatin
reviewed
Nov 17, 2025
…scripts. Still need to resolve dataset origin issue and modify buildkite yml to reflect changes
… new perf and accuracy script changes
jrplatin
reviewed
Nov 19, 2025
jrplatin
reviewed
Nov 19, 2025
jrplatin
reviewed
Nov 19, 2025
jrplatin
reviewed
Nov 19, 2025
jrplatin
reviewed
Nov 19, 2025
…nted error for multimodal inputs
JiriesKaileh
requested review from
QiliangCui,
hfan,
jcyang43,
py4,
sixiang-google,
vanbasten23,
vipannalla and
wenxindongwork
as code owners
…into jiries/llama-guard-4-text
jrplatin
approved these changes
Nov 20, 2025
jcyang43
reviewed
Nov 20, 2025
Signed-off-by: JiriesKaileh <jiries@google.com>
JiriesKaileh
force-pushed
the
jiries/llama-guard-4-text
branch
from
6fbe2ba to
c57c80a
Compare
…into jiries/llama-guard-4-text
Signed-off-by: JiriesKaileh <jiries@google.com>
jcyang43
reviewed
Nov 20, 2025
| echo -e "\n--- Running Accuracy Check (Mode: ACCURACY) ---" | ||
|
|
||
| CONFTEST_DIR="/workspace/tpu-inference/scripts/vllm/integration" | ||
| CONFTEST_DIR="/mnt/disks/jiries-disk_data/tpu-inference/scripts/vllm/integration" |
Collaborator
There was a problem hiding this comment.
Hi @JiriesKaileh, is this for your local dev environment? I don't think this works for the agents that are running the CI: link
Collaborator
Author
There was a problem hiding this comment.
Yes, line 136 should be removed. Let me open a PR for that. Thank you for point this out.
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.
Summary:
This PR introduces the JAX implementation for the text portion of the Llama-Guard-4-12B model and configures its end-to-end continuous integration (CI) pipeline. This implementation successfully parity with the PyTorch/TorchAX baseline.
Relevant bugs:
b/439655882
Performance Testing and Verification
The JAX implementation now runs consistently faster than the PyTorch baseline, achieving superior throughput and TTFT latency.
See
meta-llama_Llama-Guard-4-12B.ymlfor unit, integration, and performance tests