freeliuzc 52eda7fdb3 [Feature][MTP]support new speculative decoding method named hybrid mtp with ngram (#3610)

2025-08-26 14:29:22 +08:00

..

draft_model

[Feature][MTP]support new speculative decoding method named hybrid mtp with ngram (#3610)

2025-08-26 14:29:22 +08:00

ngram_match.cc

Sync v2.0 version of code to github repo

2025-06-29 23:29:37 +00:00

speculate_calcu_accept_ratio.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_clear_accept_nums.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_get_output_padding_offset.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_get_output.cc

[Feature][MTP]support new speculative decoding method named hybrid mtp with ngram (#3610)

2025-08-26 14:29:22 +08:00

speculate_get_padding_offset.cu

[Code Simplification] remove cum_offsets (#3410)

2025-08-18 20:21:25 +08:00

speculate_get_seq_lens_output.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_msg.h

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_save_output.cc

[Feature][MTP]support new speculative decoding method named hybrid mtp with ngram (#3610)

2025-08-26 14:29:22 +08:00

speculate_set_value_by_flags.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_step_reschedule.cu

polish code with new pre-commit rule (#2923)

2025-07-19 23:19:27 +08:00

speculate_step_system_cache.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_step.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_stop_generation_multi_stop_seqs.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_token_penalty_multi_scores.cu

[Feature] support bad_words (#3055)

2025-07-30 09:31:29 +08:00

speculate_update_input_ids_cpu.cc

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00

speculate_update.cu

[Feature][MTP]support new speculative decoding method named hybrid mtp with ngram (#3610)

2025-08-26 14:29:22 +08:00

speculate_verify.cu

[MTP] optimize mtp infer speed (#2840)

2025-07-14 19:50:22 +08:00

top_p_candidates.cu

[LLM] First commit the llm deployment code

2025-06-09 19:20:15 +08:00