原书出处
Build a Reasoning Model (From Scratch)
iterations and scoring support def self_refinement_loop( model, tokenizer, raw_prompt, device, iterations=2, max_response_tokens=2048, max_critique_tokens=256, score_fn=None, prompt_renderer=render_prompt, prompt_suffix="", verbose=False, temperature=0.7, top_p=0.9, ): steps = [] #A prompt = prompt_renderer(raw_prompt) + prompt_suffix current_full = generate_text_stream_concat_flex( model=model, tokenizer=tokenizer, prompt=prompt, device=device, max_new_tokens=max_response_tokens, verbose=False, generate_func=generate_text_top_p_stream_cache, temperature=temperature, top_p=top_p, ) current_extracted = extract_final_candidate( current_full, fallback="number_then_full" ) if score_fn: current_score = score_fn(answer=current_full, prompt=prompt) else: current_score = 0.0 #B for it in range(iterations): draft_before_full =
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-5.html
EPUB SHA-256:52d4d07986020be70f5aac21762dfcf60fcba08db0ba29d4f63957caf12aa377