[CI] Disable CUDA graphs for GLM PCP evals (#54745)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Kevin H. Luu committed
Sep 1, 2026 at 13:24 UTC
4bf06be985c64edbb0c8b0f994ceff787ce84b09
2 files changed
+2
-2
tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml
+1
-1
@@ -4,7 +4,7 @@ num_questions: 1319
4
num_fewshot: 5
5
max_concurrency: 100
6
server_args: >-
7
- --compilation-config '{"cudagraph_mode":"PIECEWISE"}'
7
+ --compilation-config '{"cudagraph_mode":"NONE"}'
8
--max-model-len 4096
9
--max-num-batched-tokens 32768
10
--safetensors-load-strategy prefetch
tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP2-PCP2-EP.yaml
+1
-1
@@ -4,7 +4,7 @@ num_questions: 1319
4
num_fewshot: 5
5
max_concurrency: 100
6
server_args: >-
7
- --compilation-config '{"cudagraph_mode":"PIECEWISE"}'
7
+ --compilation-config '{"cudagraph_mode":"NONE"}'
8
--max-model-len 4096
9
--max-num-batched-tokens 32768
10
--safetensors-load-strategy prefetch