fix summary table hardware

Author: Oseltamivir

.github/workflows/benchmark-tmpl.yml

@@ -134,6 +134,7 @@ jobs:
       - name: Launch job script
         env:
           RUNNER_NAME: ${{ runner.name }}
+          RUNNER_TYPE: ${{ inputs.runner }}
           RESULT_FILENAME: ${{ env.EXP_NAME }}_${{ env.PRECISION }}_${{ env.FRAMEWORK }}_tp${{ env.TP }}_ep${{ env.EP_SIZE }}_dpa_${{ env.DP_ATTENTION }}_conc${{ env.CONC }}_${{ runner.name }}
           # Suppress per-job eval markdown from being appended to the step summary.
           # We'll publish a single combined eval table in the collection job instead.

runners/launch_b200-dgxc.sh

@@ -41,7 +41,7 @@ docker run --rm --init --network host --name $server_name \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e ISL -e OSL -e PORT=$PORT -e EP_SIZE -e DP_ATTENTION \
 -e NCCL_GRAPH_REGISTER=0 \
 -e TORCH_CUDA_ARCH_LIST="10.0" -e CUDA_DEVICE_ORDER=PCI_BUS_ID -e CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
--e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e NUM_PROMPTS -e RUN_EVAL \
+-e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e NUM_PROMPTS -e RUN_EVAL -e RUNNER_TYPE \
 --entrypoint=/bin/bash \
 $(echo "$IMAGE" | sed 's/#/\//') \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}_docker.sh"

runners/launch_b200-nvd.sh

@@ -42,7 +42,7 @@ docker run --rm --init --network host --name $server_name \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e ISL -e OSL -e PORT=$PORT -e EP_SIZE -e DP_ATTENTION \
 -e NCCL_GRAPH_REGISTER=0 \
 -e TORCH_CUDA_ARCH_LIST="10.0" -e CUDA_DEVICE_ORDER=PCI_BUS_ID -e CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
--e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e NUM_PROMPTS -e RUN_EVAL \
+-e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e NUM_PROMPTS -e RUN_EVAL -e RUNNER_TYPE \
 --entrypoint=/bin/bash \
 $(echo "$IMAGE" | sed 's/#/\//') \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}_docker.sh"

runners/launch_h100-cr.sh

@@ -10,7 +10,7 @@ docker run --rm --network=host --name=$server_name \
 --runtime=nvidia --gpus=all --ipc=host --privileged --shm-size=16g --ulimit memlock=-1 --ulimit stack=67108864 \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
--e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e ISL -e OSL -e RUN_EVAL -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e PORT=$PORT \
+-e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e ISL -e OSL -e RUN_EVAL -e RUNNER_TYPE -e RESULT_FILENAME -e RANDOM_RANGE_RATIO -e PORT=$PORT \
 -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e TORCH_CUDA_ARCH_LIST="9.0" -e CUDA_DEVICE_ORDER=PCI_BUS_ID -e CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
 --entrypoint=/bin/bash \
 $IMAGE \

runners/launch_mi300x-amd.sh

@@ -14,7 +14,7 @@ docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT \
--e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL \
+-e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL -e RUNNER_TYPE \
 --entrypoint=/bin/bash \
 $IMAGE \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi300x_docker.sh"

runners/launch_mi300x-cr.sh

@@ -14,7 +14,7 @@ docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT \
--e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL \
+-e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL -e RUNNER_TYPE \
 --entrypoint=/bin/bash \
 $IMAGE \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi300x_docker.sh"

runners/launch_mi355x-amd.sh

@@ -36,7 +36,7 @@ docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT -e NUM_PROMPTS \
--e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL \
+-e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME -e RUN_EVAL -e RUNNER_TYPE \
 --entrypoint=/bin/bash \
 $IMAGE \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi355x_docker.sh"

utils/collect_eval_results.py

@@ -293,12 +293,9 @@ def main():
         else:
             continue
 
-        # Merge with meta
-        # Only use explicit hardware label written to meta_env.json ('hw')
-        hw_value = meta.get('hw', 'Unknown GPU')
         row = {
             'model': m.get('model') or meta.get('model') or 'unknown',
-            'hw': hw_value,
+            'hw': meta.get('hw'),
             'framework': (meta.get('framework') or 'unknown').lower(),
             'precision': (meta.get('precision') or 'unknown').lower(),
             'tp': int(meta.get('tp') or 1),