File size: 2,113 Bytes
750b88d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 | """Answering, judging and memorize prompts."""
from __future__ import annotations
ANSWER_SYS = ("You are a clinician answering a question about ONE patient. You have the patient's memory below "
"as your ONLY source. Find the relevant facts and answer.")
ANSWER_USER = ("PATIENT MEMORY (your memory):\n{ctx}\n\n"
"Today's date: {as_of}.\n\nQUESTION:\n{qtext}\n\n"
'Return JSON: {{"answer": "<your answer>"}}')
ANSWER_NOCTX = ("You are a clinician answering a question about ONE patient. The patient's "
"dialogue history is your ONLY source. Find the relevant facts and answer.\n\n"
"Today's date: {as_of}.\n\nQUESTION:\n{qtext}\n\n"
'Return JSON: {{"answer": "<your answer>"}}')
ANSWER_NOCTX_NOJSON = ("You are a clinician answering a question about ONE patient. The patient's "
"dialogue history is your ONLY source. Find the relevant facts and answer.\n\n"
"Today's date: {as_of}.\n\nQUESTION:\n{qtext}")
JUDGE_SYS = ("Judge whether CANDIDATE matches GOLD in meaning for this patient. Ignore wording/format; match "
"the core clinical fact; same MC option = match. If GOLD says the information is not documented, "
"CANDIDATE matches if it says the information is not in the record, in any wording; \"unknown\" or "
"\"cannot determine\" does not match. Return ONLY JSON.")
JUDGE_USER = 'QUESTION: {question}\nGOLD: {gold}\nCANDIDATE: {cand}\n{{"match": true|false}}'
MEMORIZE_TEMPLATE = "This is a clinical visit record. Store the information it contains in memory:\n\n{context}"
def qtext(q_raw: dict) -> str:
"""Question text, with its options appended when they are not already part of it."""
s = q_raw.get("question", "")
opts = q_raw.get("options")
if opts:
first = opts[0]
already = f"{first.get('label')}." in s and str(first.get('content', ''))[:20] in s
if not already:
s += "\n" + "\n".join(f" {o.get('label')}. {o.get('content')}" for o in opts)
return s
|