apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Inference, rename] remove padding_offsets from atten use batch_id_per_token (#2880)

Browse Source 
* remove padding_offsets from atten
This commit is contained in:
周周周
2025-07-17 18:41:31 +08:00
committed by GitHub
parent d49f8fb30a
commit ddb10ac509
50 changed files with 311 additions and 288 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_bfloat16_bfloat16_kernel.cu
View File

@@ -37,7 +37,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::bfloat16
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_bfloat16_fp8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::float8_e
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_bfloat16_int8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, int8_t>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_float16_float16_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float16>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_float16_fp8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float8_e4
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c16_float16_int8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, int8_t>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_bfloat16_bfloat16_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::bfloat16, paddle::bfloat16>
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_bfloat16_fp8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::bfloat16, paddle::float8_e4
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_bfloat16_int8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::bfloat16, int8_t>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_float16_float16_kernel.cu
View File

@@ -37,7 +37,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::float16, paddle::float16>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_float16_fp8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::float16, paddle::float8_e4m
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c4_float16_int8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC4Kernel<paddle::float16, int8_t>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_bfloat16_bfloat16_kernel.cu
View File

@@ -38,7 +38,7 @@ CascadeAppendAttentionC8Kernel<paddle::bfloat16, paddle::bfloat16, false>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -85,7 +85,7 @@ CascadeAppendAttentionC8Kernel<paddle::bfloat16, paddle::bfloat16, true>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_bfloat16_fp8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::bfloat16, paddle::float8_e4
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -80,7 +80,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::bfloat16, paddle::float8_e4
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_bfloat16_int8_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::bfloat16, int8_t, false>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -82,7 +82,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::bfloat16, int8_t, true>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_float16_float16_kernel.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, paddle::float16, f
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -82,7 +82,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, paddle::float16, t
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_float16_fp8_kerne.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, paddle::float8_e4m
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -81,7 +81,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, paddle::float8_e4m
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

4
custom_ops/gpu_ops/append_attn/template_instantiation/append_attention_c8_float16_int8_kerne.cu
View File

@@ -36,7 +36,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, int8_t, false>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,
@@ -81,7 +81,7 @@ template void CascadeAppendAttentionC8Kernel<paddle::float16, int8_t, true>(
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_table,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/encoder_write_cache_with_rope_bfloat16_bfloat16_kernel.cu
View File

@@ -22,7 +22,7 @@ EncoderWriteCacheWithRopeKernel<paddle::bfloat16, paddle::bfloat16>(
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/encoder_write_cache_with_rope_bfloat16_int_kernel.cu
View File

@@ -21,7 +21,7 @@ template void EncoderWriteCacheWithRopeKernel<paddle::bfloat16, int>(
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/encoder_write_cache_with_rope_float16_float16_kernel.cu
View File

@@ -21,7 +21,7 @@ template void EncoderWriteCacheWithRopeKernel<paddle::float16, paddle::float16>(
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const paddle::Tensor& batch_ids,

2
custom_ops/gpu_ops/append_attn/template_instantiation/encoder_write_cache_with_rope_float16_int_kernel.cu
View File

@@ -21,7 +21,7 @@ template void EncoderWriteCacheWithRopeKernel<paddle::float16, int>(
const paddle::Tensor& seq_lens_this_time,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& padding_offsets,
const paddle::Tensor& batch_id_per_token,
const paddle::Tensor& cu_seqlens_q,
const paddle::Tensor& block_tables,
const paddle::Tensor& batch_ids,