[ci] Enhance CI workflows with performance history recording

Author: orionpapadakis

.github/workflows/build-and-run.yml

@@ -242,14 +242,51 @@ jobs:
           ./llama-tornado --gpu --${{ matrix.backend.name }} \
             --model /$MODELS_DIR/granite-4.0-1b-Q8_0.gguf \
             --prompt "Say hello"
-      - name: Publish performance summary
+
+      - name: Publish summary and record history
+        if: always()
+        run: |
+          HISTORY_ARG=""
+          if [[ "${{ github.event_name }}" == "push" && "${{ github.ref }}" == "refs/heads/main" ]]; then
+            HISTORY_ARG="--history docs/perf-history.jsonl"
+          fi
+          echo "## Llama-3.2-1B-Instruct F16 — ${{ matrix.backend.name }}" >> $GITHUB_STEP_SUMMARY
+          python3 scripts/process_metrics.py \
+            --metrics     ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-standard.json:standard \
+            --metrics     ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-prefill-decode.json:prefill-decode \
+            --metrics     ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-batch-prefill-decode.json:batch-prefill-decode \
+            --backend     ${{ matrix.backend.name }} \
+            --model       Llama-3.2-1B-Instruct \
+            --quantization F16 \
+            --commit      ${{ github.sha }} \
+            --branch      ${{ github.ref_name }} \
+            --run-id      ${{ github.run_id }} \
+            --run-number  ${{ github.run_number }} \
+            --run-attempt ${{ github.run_attempt }} \
+            --workflow    "${{ github.workflow }}" \
+            $HISTORY_ARG \
+            >> $GITHUB_STEP_SUMMARY
+
+      - name: Upload metrics artifacts
         if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: metrics-${{ matrix.backend.name }}-${{ github.run_id }}
+          path: ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-*.json
+          if-no-files-found: warn
+
+      - name: Commit performance history
+        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
         run: |
-          echo "## Llama-3.2-1B FP16 — ${{ matrix.backend.name }}" >> $GITHUB_STEP_SUMMARY
-          echo "### Standard" >> $GITHUB_STEP_SUMMARY
-          cat ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-standard.md >> $GITHUB_STEP_SUMMARY
-          echo "### Prefill-Decode" >> $GITHUB_STEP_SUMMARY
-          cat ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-prefill-decode.md >> $GITHUB_STEP_SUMMARY
-          echo "### Batch-Prefill-Decode" >> $GITHUB_STEP_SUMMARY
-          cat ${{ runner.temp }}/metrics-${{ matrix.backend.name }}-batch-prefill-decode.md >> $GITHUB_STEP_SUMMARY
+          git config user.name  "github-actions[bot]"
+          git config user.email "github-actions[bot]@users.noreply.github.com"
+          git add docs/perf-history.jsonl
+          git diff --cached --quiet && echo "No history changes to commit" && exit 0
+          git commit -m "perf: record ${{ matrix.backend.name }} run ${{ github.run_id }} @ ${GITHUB_SHA::8}"
+          for attempt in 1 2 3; do
+            git pull --rebase origin main && git push && break || {
+              [ $attempt -lt 3 ] && { echo "Attempt $attempt failed, retrying in $((attempt * 5))s..."; sleep $((attempt * 5)); } \
+                                 || { echo "::error::Failed to push after 3 attempts"; exit 1; }
+            }
+          done