Source symbols: scripts/chat_sft.py
Deterministic, value-free symbol metadata extracted from public source. Implementations and literal values are not copied into this page.
- Revision: master@92d63d4e8bb4df75c3b71618f31ddde2378b2bcd · 2026-07-04T18:18:25Z
- Declarations: 5
- Referenced calls: 84
Declared symbols
sft_data_generator_bos_bestfit— def at line 180refill_buffer— def at line 204; referenceslen,render_conversation,append,range,enumerate,pop,extend,tensor,contiguous,sft_data_generator_bos_bestfit,min,next,float,all_reduce,bool,item,eval,build_val_loader,evaluate_bpb,print0get_lr_multiplier— def at line 307; referencesmin,next,float,tensor,all_reduce,bool,item,eval,build_val_loader,evaluate_bpb,print0,logget_muon_momentum— def at line 317; referencesmin,next,float,tensor,all_reduce,bool,item,eval,build_val_loader,evaluate_bpb,print0,logcentered_mean— def at line 377; referencessum,len,print0,log,items
File-level call references
ArgumentParser— line 35add_argument— line 37parse_args— line 67vars— line 68copy— line 68autodetect_device_type— line 72compute_init— line 73print0— line 75get_device_name— line 79get_peak_flops— line 80float— line 83DummyWandb— line 87init— line 87load_model— line 94get— line 97getattr— line 106setattr— line 110compile— line 118estimate_flops— line 120get_token_bytes— line 128setup_optimizer— line 132get_base_dir— line 138load_optimizer_state— line 140load_state_dict— line 143zip— line 145GradScaler— line 152SmolTalk— line 163MMLU— line 164range— line 164GSM8K— line 165TaskMixture— line 167len— line 192get_bos_token_id— line 195render_conversation— line 208append— line 209refill_buffer— line 227enumerate— line 234pop— line 242extend— line 243tensor— line 281contiguous— line 282sft_data_generator_bos_bestfit— line 300min— line 318next— line 324all_reduce— line 336bool— line 337item— line 337eval— line 341build_val_loader— line 342evaluate_bpb— line 344log— line 348train— line 354and— line 359Engine— line 361run_chat_eval— line 372sum— line 378centered_mean— line 379items— line 387join— line 394save_checkpoint— line 395state_dict— line 398get_vocab_size— line 405synchronize— line 423time— line 424model— line 426detach— line 427scale— line 430backward— line 430max— line 434get_lr_multiplier— line 436get_muon_momentum— line 437unscale_— line 443is_ddp_initialized— line 444_found_inf_per_device— line 445values— line 445step— line 447update— line 448zero_grad— line 451int— line 464collect— line 486freeze— line 487disable— line 488finish— line 498compute_cleanup— line 499
Source: scripts/chat_sft.py