fix: makefile edit for unit tests

.github/workflows/e2e-preset-test.yml

@@ -33,7 +33,7 @@ jobs:
       - name: Determine Affected Models
         id: affected_models
         run: |
-            PR_BRANCH=${{ github.head_ref }} \
+            PR_BRANCH=${{ github.ref_name }} \
             python3 .github/workflows/kind-cluster/determine_models.py
               
       - name: Print Determined Models

.github/workflows/preset-image-build.yml

@@ -118,5 +118,6 @@ jobs:
         if: ${{ always() }}
         run: |
           kubectl get job --no-headers -o custom-columns=":metadata.name" | grep "^docker-build-job-${{ matrix.model.name }}-[0-9]" | xargs -r kubectl delete job
-          kubectl get pods --no-headers -o custom-columns=":metadata.name" | grep "^docker-build-job-${{ matrix.model.name }}-[0-9]" | xargs -r kubectl delete pod
+          # Job deletion above deletes associated pods
+          # kubectl get pods --no-headers -o custom-columns=":metadata.name" | grep "^docker-build-job-${{ matrix.model.name }}-[0-9]" | xargs -r kubectl delete pod
 

.github/workflows/tests.yml

@@ -38,6 +38,10 @@ jobs:
         run: |
           make unit-test
 
+      - name: Run inference api unit tests
+        run: |
+          make inference-api-e2e
+
       - name: Upload Codecov report
         uses: codecov/codecov-action@v4
         with:

Makefile

@@ -82,6 +82,8 @@ fmt: ## Run go fmt against code.
 unit-test: ## Run unit tests.
 	go test -v $(shell go list ./pkg/... ./api/... | grep -v /vendor) -race -coverprofile=coverage.txt -covermode=atomic
 	go tool cover -func=coverage.txt
+
+inference-api-e2e: 
 	pip install -r presets/inference/text-generation/requirements.txt
 	pytest -o log_cli=true -o log_cli_level=INFO .
 

docker/presets/llama-2/Dockerfile

@@ -17,7 +17,6 @@ RUN git clone https://github.com/facebookresearch/llama
 
 WORKDIR /workspace/llama
 
-# RUN sed -i $'/torch.distributed.init_process_group("nccl")/c\\            import datetime\\\n            torch.distributed.init_process_group("nccl", timeout=datetime.timedelta(days=365*100))' /workspace/llama/llama/generation.py
 RUN ["/bin/bash", "-c", "sed -i $'/torch.distributed.init_process_group(\"nccl\")/c\\            import datetime\\\n            torch.distributed.init_process_group(\"nccl\", timeout=datetime.timedelta(days=365*100))' /workspace/llama/llama/generation.py"]
 
 RUN pip install -e .