Spaces:
Sleeping
Sleeping
File size: 2,717 Bytes
e0c069c 271fdae e0c069c |
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 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 |
import os
import logging
import requests
from time import perf_counter, sleep
from memory_manager import embed_and_store, retrieve_relevant
# Agent prompts
PROMPTS = {
"Initiator": "You are the Discussion Initiator...",
"Responder": "You are the Critical Responder...",
"Guardian": "You are the Depth Guardian...",
"Provocateur": "You are the Cross-Disciplinary Provocateur...",
"Cultural": "You are the Cultural Perspective...",
"Judge": "You are the Impartial Judge..."
}
CHAT_MODEL = os.environ.get("CHAT_MODEL", "HuggingFaceH4/zephyr-7b-beta")
HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
def safe_chat(system_prompt: str, history: list, temperature: float = 0.7) -> str:
"""Call HF inference API with timing and error handling."""
start = perf_counter()
payload = {
"inputs": [{"role": "system", "content": system_prompt}] + history,
"parameters": {"max_new_tokens": 300, "temperature": temperature}
}
headers = {"Authorization": f"Bearer {HF_API_TOKEN}"} if HF_API_TOKEN else {}
try:
resp = requests.post(
f"https://api-inference.huggingface.co/models/{CHAT_MODEL}",
json=payload,
headers=headers,
timeout=60
)
if resp.status_code == 200:
data = resp.json()
text = data[0].get('generated_text', '').strip()
elif resp.status_code == 503:
logging.warning("Model loading, retrying...")
sleep(15)
return safe_chat(system_prompt, history, temperature)
else:
logging.error(f"HF error {resp.status_code}: {resp.text}")
text = f"⚠️ API Error {resp.status_code}"
except Exception as e:
logging.error(f"safe_chat exception: {e}")
text = f"⚠️ System Error: {e}"
elapsed = perf_counter() - start
logging.info(f"safe_chat: {elapsed:.3f}s for '{system_prompt[:30]}...'")
return text
def step_turn(conversation: list, turn: int, topic: str, params: dict) -> list:
"""Advance one turn of the multi-agent conversation."""
# Choose agent by sequence
sequence = ["Initiator", "Responder", "Guardian", "Provocateur", "Cultural"]
agent = sequence[turn % len(sequence)]
prompt = PROMPTS.get(agent, "")
# Prepare history
history = [{"role": "user", "content": msg['text']} for msg in conversation[-5:]]
response = safe_chat(prompt, history, temperature=params[agent]['creativity'])
embed_and_store(response, agent, topic)
conversation.append({"agent": agent, "text": response, "turn": turn + 1})
return conversation |