Skip to content

feat(ci): Automate updating LKG commits for vllm and tpu-inference #1438

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sizhit2 wants to merge 3 commits into
base: main
Choose a base branch
from
+289 −119
Open

feat(ci): Automate updating LKG commits for vllm and tpu-inference #1438

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3a9b4b0
Automate LKG updates in nightly regression workflow
google-labs-jules[bot] Apr 28, 2026
e027f14
ci: add automated LKG capture to nightly regression tests
google-labs-jules[bot] Apr 29, 2026
de62d11
ci: automate LKG commits generation for vllm and tpu-inference
google-labs-jules[bot] Apr 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
150 changes: 150 additions & 0 deletions .github/actions/run_regression_tests/action.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,150 @@
name: "Run Regression Tests"
description: "Runs Tunix regression tests, SFT shell scripts, and RL shell scripts."
inputs:
hf_token:
description: "HuggingFace token for model downloads"
required: true

runs:
using: "composite"
steps:
- name: Run regression scripts
env:
HF_TOKEN: ${{ inputs.hf_token }}
id: regression_tests
shell: bash
run: |
# Download GSM8K dataset
mkdir -p /tmp/grpo_test/rl/grpo/data

FAILED=0
echo "📦 Executing: examples/deepscaler/math_eval_nb.py..."
python examples/deepscaler/math_eval_nb.py || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vanilla rollout engine in colocated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vanilla rollout engine in 2 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --cluster-setup=disaggregated-2-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vanilla rollout engine in 3 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --cluster-setup=disaggregated-3-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vllm rollout engine in colocated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=vllm || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vllm rollout engine in 2 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=vllm --cluster-setup=disaggregated-2-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vllm rollout engine in 3 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=vllm --cluster-setup=disaggregated-3-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vllm server mode in 2 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=vllm --rollout-server-mode=True --cluster-setup=disaggregated-2-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with vllm server mode in 3 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=vllm --rollout-server-mode=True --cluster-setup=disaggregated-3-way || FAILED=1

# SGLang Tests
unset JAX_PLATFORMS
pip list | egrep 'jax|flax|libtpu'
cd ..
git clone https://github.com/sgl-project/sglang-jax.git && cd sglang-jax/python && pip install -e . && cd ../..
pip install jax==0.8.1 flax==0.12.0 libtpu==0.0.24
pip list | egrep 'jax|flax|libtpu'
cd tunix

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with sglang_jax in colocated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=sglang_jax || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with sglang_jax in 2 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=sglang_jax --cluster-setup=disaggregated-2-way || FAILED=1

echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with sglang_jax in 3 way disaggregated mode ..."
python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine=sglang_jax --cluster-setup=disaggregated-3-way || FAILED=1

# echo "📦 Executing: scripts/grpo_demo_llama3_qwen2.py with sglang_jax with LoRA ..."
# python scripts/grpo_demo_llama3_qwen2.py --root-dir=/tmp/grpo_test --num-batches=20 --rollout-engine sglang_jax --enable-lora --lora-target-modules all || FAILED=1


if [ "$FAILED" -ne 0 ]; then
echo "One or more scripts failed!"
exit 1
fi

- name: Run SFT shell scripts
env:
HF_TOKEN: ${{ inputs.hf_token }}
shell: bash
run: |
SCRIPT_DIR="./tunix/examples/sft/mtnt"
MAX_STEPS=5
EVAL_EVERY_N_STEPS=1

# Check if directory exists
if [ ! -d "$SCRIPT_DIR" ]; then
echo "Directory $SCRIPT_DIR does not exist"
exit 1
fi

echo "🔍 Finding scripts in $SCRIPT_DIR"
for script in "$SCRIPT_DIR"/*.sh; do
if [ -f "$script" ]; then
echo "📦 Executing: $script"
echo "MAX_STEPS=$MAX_STEPS, EVAL_EVERY_N_STEPS=$EVAL_EVERY_N_STEPS"
chmod +x "$script"
if bash "$script" \
--training_config.max_steps "$MAX_STEPS" \
--training_config.eval_every_n_steps "$EVAL_EVERY_N_STEPS"; then
echo "✅ Successfully completed: $script"
else
exit_code=$?
echo "❌ Failed to complete: $script (Exit Code: $exit_code)" >&2
exit "$exit_code"
fi
fi
done
echo "🎉 All SFT scripts completed successfully."

- name: Run RL shell scripts
env:
HF_TOKEN: ${{ inputs.hf_token }}
shell: bash
run: |
SCRIPT_DIR="./tunix/examples/rl/grpo/gsm8k"
EXCLUDE_DIR="verl_compatible"
MAX_STEPS=5
EVAL_EVERY_N_STEPS=1

if [ ! -d "$SCRIPT_DIR" ]; then
echo "Directory $SCRIPT_DIR does not exist" >&2
exit 1
fi

echo "🔍 Finding scripts in $SCRIPT_DIR, excluding $EXCLUDE_DIR"
final_exit_code=0

while IFS= read -r script; do
if [ -f "$script" ]; then
echo "📦 Executing: $script"
echo "MAX_STEPS=$MAX_STEPS, EVAL_EVERY_N_STEPS=$EVAL_EVERY_N_STEPS"
chmod +x "$script"
if ! bash "$script" \
--rl_training_config.max_steps "$MAX_STEPS" \
--rl_training_config.eval_every_n_steps "$EVAL_EVERY_N_STEPS"; then
exit_code=$?
echo "❌ Failed to complete: $script (Exit Code: $exit_code)" >&2
final_exit_code=$exit_code
# Stop processing further scripts after the first failure
break
else
echo "✅ Successfully completed: $script"
fi
fi
done < <(find "$SCRIPT_DIR" -name "*.sh" -type f | grep -v "$SCRIPT_DIR/$EXCLUDE_DIR/")

if [ "$final_exit_code" -ne 0 ]; then
echo "🚫 One or more RL scripts failed. Exiting with code $final_exit_code." >&2
exit "$final_exit_code"
fi
echo "🎉 All RL scripts completed successfully."
3 changes: 2 additions & 1 deletion .github/workflows/build_and_test_tunix_nightly_regression.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,8 @@ concurrency:
cancel-in-progress: false

permissions:
contents: read
contents: write
pull-requests: write
jobs:
build_tunix_package:
name: Build tunix package
Expand Down
Loading
Loading