[PCP][DCP] Enable PCP+DCP on sparse-MLA models (#56157)

Signed-off-by: patryk.saffer <patryk.saffer@blc-login-1.blc1.mistralcompute.ai>
Signed-off-by: patryk.saffer <patryk.saffer@blc-login-0.blc1.mistralcompute.ai>
Signed-off-by: PatrykSaffer <patryk.saffer@mistral.ai>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: patryk.saffer <patryk.saffer@blc-login-1.blc1.mistralcompute.ai>
Co-authored-by: patryk.saffer <patryk.saffer@blc-login-0.blc1.mistralcompute.ai>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Patryk Saffer <PatrykSaffer@users.noreply.github.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
PatrykSaffer
2026-09-12 23:24:44 +00:00
committed by GitHub
co-authored by patryk.saffer patryk.saffer Lucas Wilkinson Patryk Saffer Claude Fable 5.1
parent 72d4d83157
commit ebe1dec2da
21 changed files with 917 additions and 138 deletions
@@ -0,0 +1,19 @@
model_name: "nvidia/GLM-5.2-NVFP4"
accuracy_threshold: 0.90
num_questions: 1319
num_fewshot: 5
max_concurrency: 100
server_args: >-
--compilation-config '{"cudagraph_mode":"PIECEWISE"}'
--max-model-len 4096
--max-num-batched-tokens 32768
--safetensors-load-strategy prefetch
--moe-backend flashinfer_cutlass
--prefill-context-parallel-size 4
--decode-context-parallel-size 4
--enable-expert-parallel
--kv-cache-dtype fp8
env:
PYTORCH_CUDA_ALLOC_CONF: "expandable_segments:True"
VLLM_LOGGING_LEVEL: "DEBUG"
VLLM_USE_V2_MODEL_RUNNER: "1"
+1
View File
@@ -1,2 +1,3 @@
GLM-5.2-NVFP4-TP2-PCP2-EP.yaml
GLM-5.2-NVFP4-TP1-PCP4-EP.yaml
GLM-5.2-NVFP4-TP1-PCP4-DCP4-EP.yaml