Skip to content

[Cherry-Pick][BugFix] redefine tmp_workspace using full tensor in append_attn(#6999) #7002

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Jiang-Jia-Jun merged 9 commits into from
Mar 25, 2026
+634 −536
Merged

[Cherry-Pick][BugFix] redefine tmp_workspace using full tensor in append_attn(#6999) #7002

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
b12713a
[Cherry-Pick][BugFix][APIServer] Enable control socket disable option…
Copilot Mar 1, 2026
f26e3de
[Cherry-Pick][BugFix] Fix AttributeError in recycle_gpu_blocks when p…
kevincheng2 Mar 2, 2026
7d76fd4
[RL] Clear Requests status of R3 (#6569)
gongshaotian Mar 2, 2026
ad1ea43
[Cherry-Pick] [BugFix] fix prefix tree updating timeout (#6615)(#6617)
liyonghua0910 Mar 3, 2026
f3991a8
[Cherry-Pick][BugFix] fix mtp_config in rl (#6595)(#6597)
Deleter-D Mar 3, 2026
7fb8af4
[BugFix][MTP] Skip empty_input_forward during dummy run (#6655)
yuanlehome Mar 5, 2026
90c6f26
redefine tmp_workspace using full tensor in append_attn
lizhenyun01 Mar 24, 2026
7cfbbd4
fix test
lizhenyun01 Mar 24, 2026
6164908
fix pre-commit
lizhenyun01 Mar 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 36 additions & 0 deletions custom_ops/gpu_ops/append_attention.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,9 @@ void AppendAttentionKernel(
const paddle::Tensor& qkv,
const paddle::Tensor& key_cache,
const paddle::Tensor& value_cache,
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& seq_lens_this_time,
Expand Down Expand Up @@ -137,6 +140,9 @@ void AppendAttentionKernel(
qkv_out,
key_cache,
value_cache,
tmp_workspace,
tmp_m,
tmp_d,
attn_mask,
cache_quant_type_str == "block_wise_fp8" ? cache_k_quant_scales
: cache_k_dequant_scales,
Expand Down Expand Up @@ -446,6 +452,9 @@ std::vector<paddle::Tensor> AppendAttention(
const paddle::Tensor& qkv,
const paddle::Tensor& key_cache,
const paddle::Tensor& value_cache,
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& seq_lens_this_time,
Expand Down Expand Up @@ -579,6 +588,9 @@ std::vector<paddle::Tensor> AppendAttention(
qkv,
key_cache,
value_cache,
tmp_workspace,
tmp_m,
tmp_d,
seq_lens_encoder,
seq_lens_decoder,
seq_lens_this_time,
Expand Down Expand Up @@ -655,6 +667,9 @@ std::vector<paddle::Tensor> AppendAttentionWithOutput(
const paddle::Tensor& qkv,
const paddle::Tensor& key_cache,
const paddle::Tensor& value_cache,
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::Tensor& seq_lens_encoder,
const paddle::Tensor& seq_lens_decoder,
const paddle::Tensor& seq_lens_this_time,
Expand Down Expand Up @@ -735,6 +750,9 @@ std::vector<paddle::Tensor> AppendAttentionWithOutput(
qkv,
key_cache,
value_cache,
tmp_workspace,
tmp_m,
tmp_d,
seq_lens_encoder,
seq_lens_decoder,
seq_lens_this_time,
Expand Down Expand Up @@ -825,6 +843,9 @@ std::vector<std::vector<int64_t>> AppendAttentionInferShape(
const std::vector<int64_t>& qkv_shape,
const std::vector<int64_t>& key_cache_shape,
const std::vector<int64_t>& value_cache_shape,
const std::vector<int64_t>& tmp_workspace_shape,
const std::vector<int64_t>& tmp_m_shape,
const std::vector<int64_t>& tmp_d_shape,
const std::vector<int64_t>& seq_lens_encoder_shape,
const std::vector<int64_t>& seq_lens_decoder_shape,
const std::vector<int64_t>& seq_lens_this_time_shape,
Expand Down Expand Up @@ -890,6 +911,9 @@ std::vector<paddle::DataType> AppendAttentionInferDtype(
const paddle::DataType& qkv_dtype,
const paddle::DataType& key_cache_dtype,
const paddle::DataType& value_cache_dtype,
const paddle::DataType& tmp_workspace_dtype,
const paddle::DataType& tmp_m_dtype,
const paddle::DataType& tmp_d_dtype,
const paddle::DataType& seq_lens_encoder_dtype,
const paddle::DataType& seq_lens_decoder_dtype,
const paddle::DataType& seq_lens_this_time_dtype,
Expand Down Expand Up @@ -975,6 +999,9 @@ std::vector<std::vector<int64_t>> AppendAttentionWithOutputInferShape(
const std::vector<int64_t>& qkv_shape,
const std::vector<int64_t>& key_cache_shape,
const std::vector<int64_t>& value_cache_shape,
const std::vector<int64_t>& tmp_workspace_shape,
const std::vector<int64_t>& tmp_m_shape,
const std::vector<int64_t>& tmp_d_shape,
const std::vector<int64_t>& seq_lens_encoder_shape,
const std::vector<int64_t>& seq_lens_decoder_shape,
const std::vector<int64_t>& seq_lens_this_time_shape,
Expand Down Expand Up @@ -1033,6 +1060,9 @@ std::vector<paddle::DataType> AppendAttentionWithOutputInferDtype(
const paddle::DataType& qkv_dtype,
const paddle::DataType& key_cache_dtype,
const paddle::DataType& value_cache_dtype,
const paddle::DataType& tmp_workspace_dtype,
const paddle::DataType& tmp_m_dtype,
const paddle::DataType& tmp_d_dtype,
const paddle::DataType& seq_lens_encoder_dtype,
const paddle::DataType& seq_lens_decoder_dtype,
const paddle::DataType& seq_lens_this_time_dtype,
Expand Down Expand Up @@ -1091,6 +1121,9 @@ PD_BUILD_STATIC_OP(append_attention)
.Inputs({"qkv",
"key_cache",
"value_cache",
"tmp_workspace",
"tmp_m",
"tmp_d",
"seq_lens_encoder",
"seq_lens_decoder",
"seq_lens_this_time",
Expand Down Expand Up @@ -1152,6 +1185,9 @@ PD_BUILD_STATIC_OP(append_attention_with_output)
.Inputs({"qkv",
"key_cache",
"value_cache",
"tmp_workspace",
"tmp_m",
"tmp_d",
"seq_lens_encoder",
"seq_lens_decoder",
"seq_lens_this_time",
Expand Down
85 changes: 53 additions & 32 deletions custom_ops/gpu_ops/append_attn/append_attention_c16_impl.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,11 +18,15 @@
template <typename T, typename OutT>
void CascadeAppendAttentionC16Kernel(
const AppendAttnMetaData& meta_data,
const paddle::Tensor& qkv, // [token_num, (num_heads + 2* kv_num_head) * head_dim]
const paddle::Tensor&
qkv, // [token_num, (num_heads + 2* kv_num_head) * head_dim]
const paddle::Tensor&
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -35,9 +39,8 @@ void CascadeAppendAttentionC16Kernel(
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand Down Expand Up @@ -99,6 +102,9 @@ void CascadeAppendAttentionC16Kernel(
qkv,
cache_k,
cache_v,
tmp_workspace,
tmp_m,
tmp_d,
attn_mask,
shift_bias,
smooth_weight,
Expand Down Expand Up @@ -127,13 +133,17 @@ void CascadeAppendAttentionC16Kernel(
})})})})})})
}

template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::bfloat16>(
template void
CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::bfloat16>(
const AppendAttnMetaData& meta_data,
const paddle::Tensor& qkv, // [token_num, num_heads, head_dim]
const paddle::Tensor&
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -146,9 +156,8 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::bfloat16
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand All @@ -174,13 +183,17 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::bfloat16
paddle::Tensor* out,
const int sliding_window);

template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::float8_e4m3fn>(
template void
CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::float8_e4m3fn>(
const AppendAttnMetaData& meta_data,
const paddle::Tensor& qkv, // [token_num, num_heads, head_dim]
const paddle::Tensor&
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -193,9 +206,8 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, paddle::float8_e
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand Down Expand Up @@ -228,6 +240,9 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, int8_t>(
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -240,9 +255,8 @@ template void CascadeAppendAttentionC16Kernel<paddle::bfloat16, int8_t>(
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand Down Expand Up @@ -272,24 +286,26 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float16>(
const AppendAttnMetaData& meta_data,
const paddle::Tensor& qkv, // [token_num, num_heads, head_dim]
const paddle::Tensor&
cache_k, // [max_block_num, num_heads, block_size, head_dim]
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
cache_v_scale, // [num_kv_heads, head_dim]
cache_k_scale, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
cache_k_zp, // [num_kv_heads, head_dim]
cache_v_scale, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
cache_v_zp, // [num_kv_heads, head_dim]
cache_k_zp, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
cache_v_zp, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand All @@ -315,13 +331,17 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float16>(
paddle::Tensor* out,
const int sliding_window);

template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float8_e4m3fn>(
template void
CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float8_e4m3fn>(
const AppendAttnMetaData& meta_data,
const paddle::Tensor& qkv, // [token_num, num_heads, head_dim]
const paddle::Tensor&
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -334,9 +354,8 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, paddle::float8_e4
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand Down Expand Up @@ -369,6 +388,9 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, int8_t>(
cache_k, // [max_block_num, num_heads, block_size, head_dim]
const paddle::Tensor&
cache_v, // [max_block_num, num_heads, head_dim, block_size]
paddle::Tensor& tmp_workspace,
paddle::Tensor& tmp_m,
paddle::Tensor& tmp_d,
const paddle::optional<paddle::Tensor>& attn_mask,
const paddle::optional<paddle::Tensor>&
cache_k_scale, // [num_kv_heads, head_dim]
Expand All @@ -381,9 +403,8 @@ template void CascadeAppendAttentionC16Kernel<paddle::float16, int8_t>(
const paddle::optional<paddle::Tensor>&
shift_bias, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>&
sinks, // [num_heads]
smooth_weight, // [num_kv_heads, head_dim]
const paddle::optional<paddle::Tensor>& sinks, // [num_heads]
const paddle::Tensor& seq_lens_q,
const paddle::Tensor& seq_lens_kv,
const paddle::Tensor& seq_lens_encoder,
Expand Down
Loading