freeliuzc 7a6c28781b [Speculative Decoding] Optimize attn_mask_offset and fix mtp bug (#7005) ...

* optimize attn_mask_offset and optimize mtp usage

* delete useless branch

* fix kernel format

* fix kernel runner

2026-03-25 01:52:06 -07:00

..

draft_model

[Speculative Decoding] refactor MTP and optimize spec-decoding postprocess (#6973)

2026-03-24 10:19:01 +08:00

build_sampling_params.cu

[Optimization][Speculative Decoding]Fuse padding sampling params (#6765)

2026-03-12 05:05:15 -07:00

naive_update_model_status.cu

[Speculative Decoding] refactor MTP and optimize spec-decoding postprocess (#6973)

2026-03-24 10:19:01 +08:00

ngram_match.cc

[Feature] GPU Memory Optimization and Retirement of V0 Scheduler (#6407)

2026-02-28 15:07:43 +08:00

speculate_calcu_accept_ratio.cu

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_clear_accept_nums.cu

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_get_output_with_topk.cc

[BugFix] replace ftok with custom_ftok in get_output/save_output ops (#6822)

2026-03-16 14:22:18 +08:00

speculate_get_output.cc

[BugFix] replace ftok with custom_ftok in get_output/save_output ops (#6822)

2026-03-16 14:22:18 +08:00

speculate_get_seq_lens_output.cu

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_get_token_penalty_multi_scores.cu

[Feature] GPU Memory Optimization and Retirement of V0 Scheduler (#6407)

2026-02-28 15:07:43 +08:00

speculate_limit_thinking_content_length.cu

[OP][Feature] 统一 limit_thinking_content_length CUDA 算子，支持回复长度限制与注入序列 (#6493)

2026-02-25 21:36:50 +08:00

speculate_logprob_utils.cu

[Speculative Decoding] Unify Spec and non-spec branch (#6685)

2026-03-10 23:58:44 -07:00

speculate_msg.h

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_preprocess.cu

[Model Runner] Deprecate not_need_stop (#6356)

2026-03-05 10:55:42 +08:00

speculate_save_output_with_topk.cc

[BugFix] replace ftok with custom_ftok in get_output/save_output ops (#6822)

2026-03-16 14:22:18 +08:00

speculate_save_output.cc

[BugFix] replace ftok with custom_ftok in get_output/save_output ops (#6822)

2026-03-16 14:22:18 +08:00

speculate_schedule_cache.cu

[Others] remove stop_nums (#6182)

2026-01-26 12:12:47 +08:00

speculate_set_stop_value_multi_seqs.cu

[Speculative Decoding] Fix speculate stop_seqs and fix accept_num in eos branch (#6825)

2026-03-12 23:48:24 -07:00

speculate_set_value_by_flags_and_idx.cu

[Feature] GPU Memory Optimization and Retirement of V0 Scheduler (#6407)

2026-02-28 15:07:43 +08:00

speculate_step_reschedule.cu

[BugFix] replace ftok with custom_ftok in get_output/save_output ops (#6822)

2026-03-16 14:22:18 +08:00

speculate_step_system_cache.cu

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_step.cu

cuda13.0, implement changes to CCCL (#6751)

2026-03-10 16:47:02 +08:00

speculate_update_input_ids_cpu.cc

init (#6642)

2026-03-04 21:55:31 +08:00

speculate_update.cu

[XPU] Fix PD + MTP (#6495)

2026-02-27 19:07:35 +08:00

speculate_verify.cu

[MTP] refactor MTP pre_process (#6358)

2026-02-09 10:47:15 +08:00

top_p_candidates.cu

[MTP] refactor MTP pre_process (#6358)

2026-02-09 10:47:15 +08:00

unified_update_model_status.cu

[Speculative Decoding] Optimize attn_mask_offset and fix mtp bug (#7005)

2026-03-25 01:52:06 -07:00

verify_draft_tokens.cu

[Speculative Decoding] Optimize attn_mask_offset and fix mtp bug (#7005)

2026-03-25 01:52:06 -07:00