[ROCm][Quantization][3/N] Refactor quark_moe w4a4 w/ oracle (#41436)

Signed-off-by: Bowen Bao <bowenbao@amd.com>
This commit is contained in:
Bowen Bao
2026-05-18 13:46:13 -04:00
committed by GitHub
parent 6859ca7615
commit a2c8fc6657
8 changed files with 224 additions and 158 deletions
@@ -0,0 +1,12 @@
model_name: "amd/Qwen3.5-35B-A3B-MXFP4"
accuracy_threshold: 0.89
tolerance: 0.03
num_questions: 1319
num_fewshot: 5
server_args: >-
--max-model-len 4096
--tensor-parallel-size 2
--gpu-memory-utilization 0.35
--moe-backend aiter
env:
VLLM_ROCM_USE_AITER: "1"
@@ -1,8 +1,10 @@
model_name: "amd/Qwen3.5-35B-A3B-MXFP4"
accuracy_threshold: 0.82
accuracy_threshold: 0.89
tolerance: 0.03
num_questions: 1319
num_fewshot: 5
server_args: >-
--max-model-len 4096
--tensor-parallel-size 2
--moe-backend emulation
--gpu-memory-utilization 0.35
+2 -1
View File
@@ -3,4 +3,5 @@ DeepSeek-R1-DP_MI325.yaml
DeepSeek-V3.2-TP_MI325.yaml
DeepSeek-V3.2-DP_MI325.yaml
Qwen3-30B-A3B-NVFP4.yaml
Qwen3.5-35B-A3B-MXFP4-TP2.yaml
Qwen3.5-35B-A3B-MXFP4-AITER-TP2.yaml
Qwen3.5-35B-A3B-MXFP4-EMU-TP2.yaml
@@ -1,2 +1,3 @@
Qwen3.5-35B-A3B-DEP2.yaml
Qwen3.5-35B-A3B-MXFP4-TP2.yaml
Qwen3.5-35B-A3B-MXFP4-AITER-TP2.yaml
Qwen3.5-35B-A3B-MXFP4-EMU-TP2.yaml