Update benchmarks.yml

Author: juliagmt-google

.github/workflows/benchmarks.yml

@@ -71,39 +71,43 @@ jobs:
         timeout-minutes: 30
         shell: bash
         run: |
+          source ../../../../../venv/bin/activate
           output=$(bash run.sh)
           echo "$output"
-      
-          # Extract metrics using Python and regex
-          python - << EOF
-            import re
-            import json
-            
-            text = \"\"\"${output}\"\"\"
-            
-            ttft_pattern = r"TTFT: ([\d.]+) ms ± ([\d.]+)%"
-            e2e_latency_pattern = r"E2E Latency: ([\d.]+) ms ± ([\d.]+)%"
-            tpot_pattern = r"TPOT: ([\d.]+) ms"
-            
-            ttft_match = re.search(ttft_pattern, text)
-            e2e_latency_match = re.search(e2e_latency_pattern, text)
-            tpot_match = re.search(tpot_pattern, text)
-            
-            metrics = {
-                "TTFT": {"value": ttft_match.group(1) if ttft_match else None, "std_dev": ttft_match.group(2) if ttft_match else None},
-                "E2E Latency": {"value": e2e_latency_match.group(1) if e2e_latency_match else None, "std_dev": e2e_latency_match.group(2) if e2e_latency_match else None},
-                "TPOT": {"value": tpot_match.group(1) if tpot_match else None},
-            }
-            
-            with open("metrics.json", "w") as f:
-                json.dump(metrics, f, indent=4)
-            
-            print(f"::set-output name=metrics::{json.dumps(metrics)}")
-            EOF
-            
-          # Copy the metrics.json file to the results directory
-          cp metrics.json xla/results/
 
+    # Extract metrics using Python and regex
+    python - << EOF
+import re
+import json
+
+text = \"\"\"${output}\"\"\"
+
+ttft_pattern = r"TTFT: ([\d.]+) ms ± ([\d.]+)%"
+e2e_latency_pattern = r"E2E Latency: ([\d.]+) ms ± ([\d.]+)%"
+tpot_pattern = r"TPOT: ([\d.]+) ms"
+
+ttft_match = re.search(ttft_pattern, text)
+e2e_latency_match = re.search(e2e_latency_pattern, text)
+tpot_match = re.search(tpot_pattern, text)
+
+metrics = {
+    "TTFT": {"value": ttft_match.group(1) if ttft_match else None, "std_dev": ttft_match.group(2) if ttft_match else None},
+    "E2E Latency": {"value": e2e_latency_match.group(1) if e2e_latency_match else None, "std_dev": e2e_latency_match.group(2) if e2e_latency_match else None},
+    "TPOT": {"value": tpot_match.group(1) if tpot_match else None},
+}
+
+with open("metrics.json", "w") as f:
+    json.dump(metrics, f, indent=4)
+
+print(f"::set-output name=metrics::{json.dumps(metrics)}")
+EOF
+
+    # Upload metrics as an artifact (optional)
+    # Create the results directory if it doesn't exist
+    mkdir -p ../../../../../results
+
+    # Copy the metrics.json file to the results directory
+    cp metrics.json ../../../../../results/
 
       - name: Wait For Connection
         uses: google-ml-infra/actions/ci_connection@main