Skip to content

Commit 2fbd2b8

Browse files
WANDY666wangzaijun
WANDY666
and
wangzaijun
authored
add diverse_stage2 add optimize diverse_stage1 (#1174)
Co-authored-by: wangzaijun <[email protected]>
1 parent f0481a8 commit 2fbd2b8

File tree

77 files changed

+1160
-189
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

77 files changed

+1160
-189
lines changed

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_GeForce_RTX_4090_D.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_GeForce_RTX_5090.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "128": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_GeForce_RTX_4090_D.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_GeForce_RTX_5090.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=4,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_H200.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 1}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "128": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 5}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=8,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_GeForce_RTX_4090_D.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 3}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}}, "8192": {"8": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=8,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_GeForce_RTX_5090.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}}, "8192": {"8": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 10}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 5}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=8,out_dtype=torch.bfloat16,q_head_dim=128}_NVIDIA_H200.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 64, "num_warps": 8, "num_stages": 1}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}}, "8192": {"8": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 64, "num_warps": 4, "num_stages": 2}}}

‎lightllm/common/all_kernel_configs/_fwd_kernel_flash_decode_diverse_stage1:v2/{block_seq=256,gqa_group_size=16,max_batch_group_size=8,out_dtype=torch.float16,q_head_dim=128}_NVIDIA_GeForce_RTX_4090_D.json‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
{"4096": {"8": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 3}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}}, "8192": {"8": {"BLOCK_N": 32, "num_warps": 4, "num_stages": 2}, "32": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 3}, "128": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}, "256": {"BLOCK_N": 16, "num_warps": 4, "num_stages": 2}}}

0 commit comments

Comments
 (0)