[Speculative decoding] Add periodic log with time spent in proposal/scoring/verification (#6963)

This commit is contained in:
Cade Daniel
2024-08-05 01:46:44 -07:00
committed by GitHub
parent c0d8f1636c
commit 82a1b1a82b
5 changed files with 125 additions and 35 deletions

View File

@@ -792,6 +792,7 @@ class EngineArgs:
speculative_max_model_len=self.speculative_max_model_len,
enable_chunked_prefill=self.enable_chunked_prefill,
use_v2_block_manager=self.use_v2_block_manager,
disable_log_stats=self.disable_log_stats,
ngram_prompt_lookup_max=self.ngram_prompt_lookup_max,
ngram_prompt_lookup_min=self.ngram_prompt_lookup_min,
draft_token_acceptance_method=\