apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity
Files
29a313a4022bfde8d482f9c03113b8abf866d8df
FastDeploy/custom_ops/gpu_ops
T
History
chen 29a313a402 [Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354) 
* support FA4 sm100

* flash attn backend support mask

* flash attn backend run flashmask correct

* add test for flash_attn_backend and flash_attn_func

* check

* add test for fa4

* requirements.txt add fa4 whl

* check test on sm100

* fix CI conflict

* add enable_torch_proxy for flash_mask

* lazy import fa4

* check

* fix tests import

* check test_load_mpt import
2026-02-05 14:39:00 +08:00
..
append_attn
[Model Runner] Prepare token count and move FA3 initialization into the graph (#6170)
2026-01-26 12:16:57 +08:00
common
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
custom_all_reduce
Support setting communication groups in custom_allreduce and the all-to-all\transpose fused operator during the decoding phase. (#5917)
2026-01-12 14:09:39 +08:00
cutlass_extensions
fix typos (#3684)
2025-09-01 17:50:17 +08:00
cutlass_kernels
[Feature] Support fusedmoe on Blackwell (#5325)
2025-12-16 11:58:50 +08:00
flash_mask_attn
remove unneeded para from flash_mask_attention (#6218)
2026-01-27 14:04:27 +08:00
fp8_gemm_with_cutlass
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
glog
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
int8_gemm_with_cutlass
fix typos (#3684)
2025-09-01 17:50:17 +08:00
machete
[BugFix] Fix wint4 ep issue caused by empty run (#5870)
2026-01-05 14:24:37 +08:00
mla_attn
[BugFix] fix BatchMLAWithPagedKVCacheKernel O_tmp (#5895)
2026-01-06 15:39:06 +08:00
moba_attn
enhance set_stop_value_multi_ends and standardize the registration of some operators (#4525)
2025-10-21 22:06:06 +08:00
moe
[Feature] Support Ernie FP8 on sm100 ( the fixed version) (#6304)
2026-02-03 17:47:38 +08:00
quantization
[OPs] MoE support wfp8afp8(channelwise) and improve per_token_quant_fp8 (#4238)
2025-09-24 16:39:51 +08:00
sample_kernels
[Metax] refactor cutlass moe and optimize flash attention (#5361)
2025-12-10 17:15:17 +08:00
speculate_decoding
remove speculate_get_padding_offset op (#6308)
2026-02-03 15:18:12 +08:00
w4afp8_gemm
[BugFix] fix wint2 (#6109)
2026-01-20 21:46:21 +08:00
wfp8afp8_sparse_gemm
【New Feature】支持Fp8 group Gemm 24稀疏 (#3463)
2025-08-19 02:54:47 -07:00
append_attention.cu
[Others]get_block_shape_and_split_kv_block clean code (#5123)
2025-11-20 16:40:04 +08:00
beam_search_softmax.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
cpp_extensions.cc
[Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354)
2026-02-05 14:39:00 +08:00
cuda_multiprocess.h
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
dequant_int8.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
enforce_generation.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
env.h
【Inference Optimize】DeepSeek-V3-model MLA Optimize (#3886)
2025-09-11 10:46:09 +08:00
fused_get_rotary_embedding.cu
rename fused_get_rope.cu (#3752)
2025-09-03 10:54:34 +08:00
fused_hadamard_quant_fp8.cu
[Optimize] Optimize tensorwise fp8 performance (#2729)
2025-07-07 20:06:28 +08:00
fused_mask_swiglu_fp8_quant_kernel.cu
[Optimize] optimize mask_quant & swiglu (#6222)
2026-02-02 13:52:38 +08:00
fused_neox_rope_embedding.cu
[Feature] Optim PaddleOCR-VL (#4873)
2025-11-07 14:56:44 +08:00
fused_rotary_position_encoding.cu
[Feature] DeepseekV3 use pd_build_static_op (#2948)
2025-07-22 15:03:41 +08:00
gather_idx.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
gelu_tanh.cu
[Metax] adapt to the latest develop (#6282)
2026-01-29 23:21:20 -08:00
get_attn_mask_q.cu
[Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354)
2026-02-05 14:39:00 +08:00
get_data_ptr_ipc.cu
Sync v2.0 version of code to github repo
2025-06-29 23:29:37 +00:00
get_img_boundaries.cc
[Models] Add Qwen3-VL Model Support (#5763)
2025-12-29 17:39:33 +08:00
get_mm_split_fuse.cc
enhance set_stop_value_multi_ends and standardize the registration of some operators (#4525)
2025-10-21 22:06:06 +08:00
get_output_ep.cc
[PD Disaggregation] Distinguish the pipelines for sending kv signal in different prefill (#5514)
2025-12-12 14:05:36 +08:00
get_output_msg_with_topk.cc
ep support logprob (#4089) (#4151)
2025-09-19 14:07:31 +08:00
get_output.cc
[PD Disaggregation] support DP via v1 router and decouple DP and EP (#5197)
2025-12-04 15:38:43 +08:00
get_padding_offset_system.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
get_padding_offset.cu
remove speculate_get_padding_offset op (#6308)
2026-02-03 15:18:12 +08:00
get_position_ids_and_mask_encoder_batch.cu
[Feature] DeepseekV3 use pd_build_static_op (#2948)
2025-07-22 15:03:41 +08:00
helper.cu
[Feature][OP] Append Attn Support CUDA-PDL (#5072)
2025-11-17 20:47:33 +08:00
helper.h
remove unneeded para from flash_mask_attention (#6218)
2026-01-27 14:04:27 +08:00
init_signal_layerwise.cc
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
ipc_sent_key_value_cache_by_remote_ptr.cu
[Feature] support rdma pd dy-c8 (#5788)
2026-01-07 14:55:25 +08:00
limit_thinking_content_length_v1.cu
[BugFix] Fix limit_thinking early return logic in CUDA kernels (#5471)
2025-12-10 11:03:19 +08:00
limit_thinking_content_length_v2.cu
[BugFix] Fix limit_thinking early return logic in CUDA kernels (#5471)
2025-12-10 11:03:19 +08:00
merge_prefill_decode_output.cu
集中式支持fa3 (#3112)
2025-08-01 18:03:36 +08:00
msg_utils.h
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
multi_head_latent_attention.cu
【Hackathon 9th No.86】autogen MoeFastHardamardImplWrapper template_instantiation (#4592)
2025-10-30 10:28:36 +08:00
ngram_mask.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
noaux_tc_redundant.cu
[Feature] Support noaux for eplb (#5143)
2025-11-21 14:10:32 +08:00
noaux_tc.cu
Fix noaux_tc cuda Error 700 in CUDAGraph (#4174)
2025-09-23 18:41:33 +08:00
noauxtc_kernel.h
[Others] clean code (#5691)
2025-12-24 11:28:47 +08:00
open_shm_and_get_meta_signal.cc
Sync v2.0 version of code to github repo
2025-06-29 23:29:37 +00:00
per_token_quant_fp8.cu
[Feature] FD_USE_PHI_FP8_QUANT (#6320)
2026-02-03 22:33:03 -08:00
read_data_ipc.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
read_ids.py
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
read_temp_ids.py
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
reasoning_phase_token_constraint.cu
[Others] Support constrained decoding when enable_thinking is false (#6248)
2026-01-28 00:05:17 -08:00
rebuild_padding.cu
add PADDLE_ENFORCE (#6321)
2026-02-04 10:47:19 +08:00
recover_decode_task.cu
support mtp in v1_scheduler mode (#3695)
2025-09-04 17:39:59 +08:00
remote_cache_kv_ipc.cc
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
remote_cache_kv_ipc.h
[PD Disaggregation] Distinguish the pipelines for sending kv signal in different prefill (#5514)
2025-12-12 14:05:36 +08:00
save_output_msg_with_topk.cc
[Optim] Robust sync status when preempted happens (#5796)
2026-01-14 12:07:33 +08:00
save_with_output_msg.cc
[Optim] Robust sync status when preempted happens (#5796)
2026-01-14 12:07:33 +08:00
save_with_output_msg.h
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
save_with_output.cc
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
scaled_gemm_f8_i4_f16_gemm.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
scaled_gemm_f8_i4_f16_weight_quantize.cu
fix scaled_gemm_f8_i4_f16_weight_quantize input (#3685)
2025-08-29 11:04:04 +08:00
seqs2seqs.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
set_data_ipc.cu
[BugFix] fix shm opened but not closed in set_data_ipc (#5826)
2025-12-29 23:35:07 +08:00
set_flags.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
set_mask_value.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
set_stop.cu
[Model Runner] Refactor execute_model for GPU async scheduling (#6176)
2026-01-28 14:19:33 +08:00
set_value_by_flags_and_idx.cu
Optimizing the performance of think length limit using custom operators (#4279)
2025-10-20 21:09:13 +08:00
share_external_data.cu
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
step_reschedule.cu
fix typos (#3684)
2025-09-01 17:50:17 +08:00
step_system_cache.cu
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
step.cu
Adapt for iluvatar gpu (#2684)
2025-07-07 16:53:14 +08:00
stop_generation_multi_ends.cu
[Feature] support stop_token_ids (#5399)
2025-12-09 17:49:12 +08:00
stop_generation.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
swap_cache_batch.cu
[Feature] dyc8 support prefixcache (#5125)
2025-11-21 19:46:26 +08:00
swap_cache_layout.cu
[CI] Add unit test for swap_layout && remove unit test of splitwise_scheduler (#6250)
2026-01-28 19:20:20 +08:00
swap_cache.cu
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
system2group.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
text_image_gather_scatter.cu
[Graph Optimization] Fix IR graph dependency error exposed after enabling SOT by updating the return value of TextImageGatherScatter (#4610)
2025-10-28 18:31:23 +08:00
text_image_index_out.cu
[Feat] ernie4_5_vl_moe support CudaGraph (#3226)
2025-09-10 13:11:57 +08:00
token_penalty_multi_scores.cu
[BugFix] Fix token_penalty kernel (#6069)
2026-01-28 12:03:05 +08:00
token_penalty_only_once.cu
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
token_transfer.hpp
fix typos (#3684)
2025-09-01 17:50:17 +08:00
transfer_output.cc
[LLM] First commit the llm deployment code
2025-06-09 19:20:15 +08:00
tune_cublaslt_gemm.cu
polish code with new pre-commit rule (#2923)
2025-07-19 23:19:27 +08:00
unset_data_ipc.cu
[feat] support prefix cache clearing when /clear_load_weight is called (#4008)
2025-09-28 19:42:53 +08:00
update_attn_mask_offsets.cu
[BugFix]Fix attention mask bug in D-Node of PD-split mode (#5245)
2025-11-26 17:56:28 +08:00
update_inputs_beam.cu
[XPU] bind some OPs for VL model with pybind (#4522)
2025-10-27 10:50:08 +08:00
update_inputs_v1.cu
[Model Runner] Refactor execute_model for GPU async scheduling (#6176)
2026-01-28 14:19:33 +08:00
update_inputs.cu
[Model Runner] Refactor execute_model for GPU async scheduling (#6176)
2026-01-28 14:19:33 +08:00
update_split_fuse_input.cu
fix typos (#3684)
2025-09-01 17:50:17 +08:00