chore: add new mi325x SLURM nodes (#224)

* add new mi325x scripts and configs

* adding ability to filter runner node on runner-model-sweep in god file

* use hf instead of huggingface-cli; add debug info

* no cpus per task

* 256 cpus per task

* revert erroneous change

* get rid of debugging

* whitespace

Author: cquil11

.github/configs/runners.yaml

@@ -34,6 +34,9 @@ mi300x:
 - 'mi300x-oci_0'
 mi325x:
 - 'mi325x-amd_0'
+- 'mi325x-amd_1'
+- 'mi325x-amd_2'
+- 'mi325x-amd_3'
 - 'mi325x-tw_0'
 - 'mi325x-tw_1'
 - 'mi325x-tw_2'

benchmarks/dsr1_fp8_h200_slurm.sh

@@ -17,7 +17,7 @@
 echo "JOB \$SLURM_JOB_ID running on \$SLURMD_NODENAME"
 
 pip3 install --user sentencepiece
-huggingface-cli download $MODEL
+hf download $MODEL
 PORT=$(( 8888 + $PORT_OFFSET ))
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 

benchmarks/dsr1_fp8_mi300x_slurm.sh

@@ -15,7 +15,7 @@
 
 echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
 
-huggingface-cli download $MODEL
+hf download $MODEL
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 PORT=8888

benchmarks/dsr1_fp8_mi325x_slurm.sh

@@ -4,7 +4,7 @@ echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 PORT=8888
-huggingface-cli download $MODEL
+hf download $MODEL
 
 # Reference
 # https://rocm.docs.amd.com/en/docs-7.0-rc1/preview/benchmark-docker/inference-sglang-deepseek-r1-fp8.html#run-the-inference-benchmark

benchmarks/gptoss_fp4_mi300x_slurm.sh

@@ -15,7 +15,7 @@
 
 echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
 
-huggingface-cli download $MODEL
+hf download $MODEL
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 PORT=8888

benchmarks/gptoss_fp4_mi325x_slurm.sh

@@ -16,7 +16,7 @@
 
 echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
 
-huggingface-cli download $MODEL
+hf download $MODEL
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 PORT=8888