Skip to content

Commit cf9b71c

Browse files
seungrokjclaude
andcommitted
[AMD] [AGENTX] dsv4 mi355x: force lm-eval on sglang-mtp recipe; drop runner node exclude
- dsv4_fp4_mi355x_sglang_mtp.sh: export EVAL_FRAMEWORK=lm-eval (never swebench). - dsv4_fp4_mi355x_vllm.sh: revert the EVAL_FRAMEWORK line (moved to sglang-mtp). - launch_mi355x-amds.sh: remove --exclude=mia1-p01-g09,mia1-p01-g11 from salloc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1 parent f766b5f commit cf9b71c

3 files changed

Lines changed: 8 additions & 8 deletions

File tree

benchmarks/single_node/agentic/dsv4_fp4_mi355x_sglang_mtp.sh

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -14,6 +14,13 @@ set -x
1414

1515
source "$(dirname "$0")/../../benchmark_lib.sh"
1616

17+
# Force the eval framework to lm-eval for this recipe. run_eval derives its
18+
# default as swebench for agentic scenarios (scenario_default=swebench when
19+
# IS_AGENTIC/SCENARIO_TYPE=agentic-coding), but EVAL_FRAMEWORK takes precedence
20+
# over that default (benchmark_lib.sh: framework=${EVAL_FRAMEWORK:-...}), so
21+
# setting it here makes the effective framework always lm-eval, never swebench.
22+
export EVAL_FRAMEWORK="lm-eval"
23+
1724
check_env_vars MODEL TP CONC KV_OFFLOADING TOTAL_CPU_DRAM_GB RESULT_DIR DURATION EP_SIZE DP_ATTENTION
1825

1926
if [[ -n "$SLURM_JOB_ID" ]]; then

benchmarks/single_node/agentic/dsv4_fp4_mi355x_vllm.sh

Lines changed: 0 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -29,13 +29,6 @@ set -x
2929

3030
source "$(dirname "$0")/../../benchmark_lib.sh"
3131

32-
# Force the eval framework to lm-eval for this recipe. run_eval derives its
33-
# default as swebench for agentic scenarios (scenario_default=swebench when
34-
# IS_AGENTIC/SCENARIO_TYPE=agentic-coding), but EVAL_FRAMEWORK takes precedence
35-
# over that default (benchmark_lib.sh: framework=${EVAL_FRAMEWORK:-...}), so
36-
# setting it here makes the effective framework always lm-eval, never swebench.
37-
export EVAL_FRAMEWORK="lm-eval"
38-
3932
check_env_vars MODEL TP CONC KV_OFFLOADING TOTAL_CPU_DRAM_GB RESULT_DIR DURATION EP_SIZE DP_ATTENTION
4033

4134
if [[ -n "${SLURM_JOB_ID:-}" ]]; then

runners/launch_mi355x-amds.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -268,7 +268,7 @@ else
268268
export GPU_COUNT="${GPU_COUNT:-${TP:?TP must be set}}"
269269

270270
set -x
271-
salloc --partition=$PARTITION --gres=gpu:$GPU_COUNT --exclusive --cpus-per-task=128 --time=500 --no-shell --job-name="$RUNNER_NAME" --exclude=mia1-p01-g09,mia1-p01-g11
271+
salloc --partition=$PARTITION --gres=gpu:$GPU_COUNT --exclusive --cpus-per-task=128 --time=500 --no-shell --job-name="$RUNNER_NAME"
272272
JOB_ID=$(squeue --name="$RUNNER_NAME" -h -o %A | head -n1)
273273

274274
srun --jobid=$JOB_ID bash -c "docker stop \$(docker ps -a -q)"

0 commit comments

Comments
 (0)