[Models]support blas and moe nn layout of deepseek-v3

89d1dd57 · zhuwenwen · 53076d70 · 89d1dd57 · 89d1dd57 · 89d1dd57
Commit 89d1dd57 authored Mar 25, 2025 by zhuwenwen
12 changed files
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=BW3000.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=BW3000.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=DCU_K100_AI_nn.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=DCU_K100_AI_nn.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=K100_AI.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=K100_AI.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=K100_AI_nn.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=2048,device_name=K100_AI_nn.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW200.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW200.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW200_nn.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW200_nn.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW3000.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=BW3000.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=K100_AI.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=K100_AI.json
--- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=K100_AI.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=K100_AI.json
--- a/vllm/model_executor/layers/fused_moe/fused_moe.py
+++ b/vllm/model_executor/layers/fused_moe/fused_moe.py
--- a/vllm/model_executor/layers/fused_moe/layer.py
+++ b/vllm/model_executor/layers/fused_moe/layer.py
--- a/vllm/model_executor/models/deepseek_v2.py
+++ b/vllm/model_executor/models/deepseek_v2.py