fix configs

wangzaijun · wangzaijun · commit 809829d6f7cc · 2025-11-24T09:39:29.000Z
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=16,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=16,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 9}, "32": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "32": {"BLOCK_N": 32, "num_warps": 8, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "256": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 11}}}
+{"4096": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=16,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=16,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 9}, "32": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "32": {"BLOCK_N": 32, "num_warps": 8, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "256": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 11}}}
+{"4096": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=4,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=4,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 32, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 7}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "256": {"BLOCK_N": 32, "num_warps": 2, "num_stages": 11}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 11}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=4,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=4,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 32, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 7}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "256": {"BLOCK_N": 32, "num_warps": 2, "num_stages": 11}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 11}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=5,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=5,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 4}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 8, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "128": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 10}, "256": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 1}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=5,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=5,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 4}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 8, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "128": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 10}, "256": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 1}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=8,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=8,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 16, "num_warps": 8, "num_stages": 9}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "256": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 2}}, "8192": {"8": {"BLOCK_N": 32, "num_warps": 2, "num_stages": 7}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 11}, "256": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}
diff --git a/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=8,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json b/lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v1/{block_seq=256,gqa_group_size=8,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json
@@ -1 +1 @@
-{"4096": {"8": {"BLOCK_N": 16, "num_warps": 8, "num_stages": 9}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "256": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 2}}, "8192": {"8": {"BLOCK_N": 32, "num_warps": 2, "num_stages": 7}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 11}, "256": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}}}
+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}

Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-{"4096": {"8": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 9}, "32": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "32": {"BLOCK_N": 32, "num_warps": 8, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "256": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 11}}}`
	`1`	`+{"4096": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 9}}}`
Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-{"4096": {"8": {"BLOCK_N": 32, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 16, "num_warps": 16, "num_stages": 7}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "256": {"BLOCK_N": 32, "num_warps": 2, "num_stages": 11}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "128": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 9}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 11}}}`
	`1`	`+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 4}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}`
Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-{"4096": {"8": {"BLOCK_N": 64, "num_warps": 16, "num_stages": 7}, "32": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 3}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 4}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 8, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 11}, "128": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 10}, "256": {"BLOCK_N": 64, "num_warps": 2, "num_stages": 1}}}`
	`1`	`+{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "128": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 2, "num_stages": 5}}}`