Pre release 0.4.0

.github/workflows/docker.yaml

@@ -21,7 +21,7 @@ env:
 jobs:
   build-and-push-image:
     runs-on: self-hosted
-    timeout-minutes: 240  # wait up to 4 hours
+    timeout-minutes: 480  # wait up to 8 hours
     # Sets the permissions granted to the `GITHUB_TOKEN` for the actions in this job.
     permissions:
       contents: read
@@ -61,8 +61,8 @@ jobs:
         with:
           context: trinity-${{ github.run_id }}
           push: true
-          file: trinity-${{ github.run_id }}/scripts/docker/Dockerfile
-          shm-size: 64g
+          file: trinity-${{ github.run_id }}/scripts/docker/Dockerfile.uv
+          shm-size: 128g
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
 

.github/workflows/docker/docker-compose.yaml

@@ -1,6 +1,6 @@
 services:
   trinity-node-1:
-    image: trinity-rft-unittest:20251030
+    image: trinity-rft-unittest:20251225
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --head --dashboard-host 0.0.0.0 --include-dashboard true --block"
     environment:
@@ -29,7 +29,7 @@ services:
             capabilities: [gpu]
 
   trinity-node-2:
-    image: trinity-rft-unittest:20251030
+    image: trinity-rft-unittest:20251225
     pull_policy: never
     command: sh -c "pip install -e .[dev] && ray start --address=trinity-node-1:6379 --block"
     environment:

.github/workflows/pre-commit.yaml

@@ -9,5 +9,5 @@ jobs:
     - uses: actions/checkout@v4
     - uses: actions/setup-python@v4
       with:
-        python-version: '3.10'
+        python-version: '3.12'
     - uses: pre-commit/[email protected]

.github/workflows/unittest.yaml

@@ -34,7 +34,7 @@ jobs:
         MAX_RETRIES=20
         RETRY_INTERVAL=5
         for i in $(seq 1 $MAX_RETRIES); do
-          docker compose exec trinity-node-1 ray status && docker compose exec trinity-node-2 ray status && break
+          docker compose exec trinity-node-1 bash -c "source /opt/venv/bin/activate && ray status" && docker compose exec trinity-node-2 bash -c "source /opt/venv/bin/activate && ray status" && break
           echo "Waiting for ray cluster to be ready... ($i/$MAX_RETRIES)"
           sleep $RETRY_INTERVAL
           if [ "$i" -eq "$MAX_RETRIES" ]; then
@@ -76,12 +76,12 @@ jobs:
         TYPE="${{ steps.test_type.outputs.type }}"
         if [ "$TYPE" = "all" ]; then
           echo "tests_run=true" >> $GITHUB_ENV
-          docker compose exec trinity-node-1 pytest tests -v -s --ctrf report.json
+          docker compose exec trinity-node-1 bash -c "source /opt/venv/bin/activate && pytest tests -v -s --ctrf report.json"
         elif [ "$TYPE" = "diff" ]; then
           if [ -s ../../../test_dirs.txt ]; then
             echo "tests_run=true" >> $GITHUB_ENV
             TEST_DIRS=$(cat ../../../test_dirs.txt | xargs)
-            docker compose exec trinity-node-1 pytest $TEST_DIRS -v -s --ctrf report.json
+            docker compose exec trinity-node-1 bash -c "source /opt/venv/bin/activate && pytest $TEST_DIRS -v -s --ctrf report.json"
           else
             echo "No changed modules detected, skipping tests."
             echo "tests_run=false" >> $GITHUB_ENV
@@ -90,7 +90,7 @@ jobs:
           MODULE="${{ steps.test_type.outputs.module }}"
           if [ -n "$MODULE" ]; then
             echo "tests_run=true" >> $GITHUB_ENV
-            docker compose exec trinity-node-1 pytest tests/$MODULE -v -s --ctrf report.json
+            docker compose exec trinity-node-1 bash -c "source /opt/venv/bin/activate && pytest tests/$MODULE -v -s --ctrf report.json"
           else
             echo "No module specified, skipping tests."
             echo "tests_run=false" >> $GITHUB_ENV

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name =  "trinity-rft"
-version = "0.3.3"
+version = "0.4.0"
 authors = [
     {name="Trinity-RFT Team", email="[email protected]"},
 ]
@@ -73,6 +73,8 @@ dev = [
 ]
 megatron = [
     "megatron-core[mlm]==0.13.1",
+    # if you found "undefined symbol" error in transformer engine
+    # reinstall it with --no-build-isolation and `--no-cache-dir` flag
     "transformer_engine[pytorch]==2.8.0",
     "mbridge>=0.13.0",
 ]

scripts/docker/Dockerfile.uv

@@ -39,8 +39,10 @@ RUN . /opt/venv/bin/activate && \
 
 # Install flash_attn and Megatron
 RUN . /opt/venv/bin/activate && \
-    uv pip install flash_attn==2.8.1 --no-cache-dir && \
-    uv pip install -e .[megatron] && \
+    uv pip install flash_attn==2.8.1 --no-build-isolation && \
+    uv pip install megatron-core[mlm]==0.13.1 && \
+    uv pip install transformer_engine[pytorch]==2.8.0 --no-build-isolation && \
+    uv pip install mbridge>=0.13.0 && \
     NVCC_APPEND_FLAGS="--threads 4" APEX_PARALLEL_BUILD=8 \
     uv pip install -v --no-build-isolation \
     --config-settings="--build-option=--cpp_ext" \

tests/utils/swanlab_test.py

@@ -5,6 +5,9 @@
 class TestSwanlabMonitor(unittest.TestCase):
     @classmethod
     def setUpClass(cls):
+        cls._original_env = {
+            "SWANLAB_API_KEY": os.environ.get("SWANLAB_API_KEY"),
+        }
         os.environ["SWANLAB_API_KEY"] = "xxxxxxxxxxxxxxxxxxxxx"
 
     @classmethod
@@ -31,7 +34,3 @@ def test_swanlab_monitor_smoke(self):
         # Log a minimal metric to verify basic flow
         mon.log({"smoke/metric": 1.0}, step=1)
         mon.close()
-
-
-if __name__ == "__main__":
-    unittest.main()

trinity/__init__.py

@@ -1,4 +1,4 @@
 # -*- coding: utf-8 -*-
 """Trinity-RFT (Reinforcement Fine-Tuning)"""
 
-__version__ = "0.3.3"
+__version__ = "0.4.0"

trinity/common/models/vllm_model.py

@@ -447,7 +447,7 @@ async def convert_messages_to_experience(
             if len(token_ids) > self.config.max_model_len - 1:
                 truncate_status = "response_truncated"
                 self.logger.warning(
-                    f"Warning: {len(token_ids) = } exceeds the length limit {self.config.max_model_len-1 = }"
+                    f"Warning: {len(token_ids)=} exceeds the length limit {(self.config.max_model_len - 1)=}"
                 )
                 token_ids = token_ids[: self.config.max_model_len - 1]
                 action_mask = action_mask[: self.config.max_model_len - 1]