Merge branch 'main' into evals-on-refactor

Author: cquil11

.github/configs/nvidia-master.yaml

@@ -167,34 +167,38 @@ dsr1-fp8-h200-trt:
     - { tp: 8, ep: 8, dp-attn: true, conc-start: 64, conc-end: 64 }
 
 gptoss-fp4-b200-trt:
-  image: nvcr.io#nvidia/tensorrt-llm/release:1.2.0rc0.post1
+  image: nvcr.io#nvidia/tensorrt-llm/release:1.2.0rc2
   model: openai/gpt-oss-120b
   model-prefix: gptoss
   runner: b200-trt
   precision: fp4
   framework: trt
-  # For all sequence lengths, if CONC >= 256, then EP=TP and DP_ATTN=true
+  # Enable DP_ATTENTION for conc >= 32
   seq-len-configs:
   - isl: 1024
     osl: 1024
     search-space:
+    - { tp: 2, dp-attn: true, conc-start: 32, conc-end: 128 }
+    - { tp: 4, dp-attn: true, conc-start: 32, conc-end: 64 }
     - { tp: 1, conc-start: 64, conc-end: 128 }
-    - { tp: 2, conc-start: 4, conc-end: 128 }
-    - { tp: 4, conc-start: 4, conc-end: 128 }
+    - { tp: 2, conc-start: 4, conc-end: 32 }
+    - { tp: 4, conc-start: 4, conc-end: 64 }
     - { tp: 8, conc-start: 4, conc-end: 8 }
   - isl: 1024
     osl: 8192
     search-space:
     - { tp: 1, conc-start: 64, conc-end: 128 }
+    - { tp: 2, dp-attn: true, conc-start: 64, conc-end: 128 }
     - { tp: 2, conc-start: 4, conc-end: 128 }
     - { tp: 4, conc-start: 4, conc-end: 128 }
     - { tp: 8, conc-start: 4, conc-end: 16 }
   - isl: 8192
     osl: 1024
     search-space:
     - { tp: 1, conc-start: 64, conc-end: 128 }
+    - { tp: 2, dp-attn: true, conc-start: 64, conc-end: 128 }
     - { tp: 2, conc-start: 4, conc-end: 128 }
-    - { tp: 4, conc-start: 4, conc-end: 128 }
+    - { tp: 4, conc-start: 4, conc-end: 32 }
     - { tp: 8, conc-start: 4, conc-end: 8 }
 
 gptoss-fp4-b200-vllm:

.github/workflows/label-validation.yml

@@ -6,17 +6,20 @@ concurrency:
   cancel-in-progress: true
 
 on:
-  pull_request:
+  pull_request_target:
     types: [labeled, synchronize]
 
 jobs:
   get-jobs:
     runs-on: ubuntu-latest
+    environment: fork-pr-validation
     outputs:
       search-space-config: ${{ steps.get-jobs.outputs.search-space-config }}
     steps:
       - name: Checkout code
         uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
 
       - id: get-jobs
         shell: python

benchmarks/gptoss_fp4_b200_trt_slurm.sh

@@ -31,7 +31,6 @@ echo "MOE_BACKEND set to '$MOE_BACKEND'"
 
 EXTRA_CONFIG_FILE="gptoss-fp4.yml"
 export TRTLLM_ENABLE_PDL=1
-export NCCL_GRAPH_REGISTER=0
 
 cat > $EXTRA_CONFIG_FILE << EOF
 cuda_graph_config:
@@ -50,6 +49,9 @@ moe_config:
 EOF
 
 if [[ "$DP_ATTENTION" == "true" ]]; then
+    export TRTLLM_MOE_ALLTOALL_BACKEND="mnnvlthroughput"
+    export TRTLLM_FORCE_ALLTOALL_METHOD="MNNVL"
+    export TRTLLM_MOE_A2A_WORKSPACE_MB="2048"
     cat << EOF >> $EXTRA_CONFIG_FILE
 attention_dp_config:
     enable_balance: true