Spaces:
Running
Running
Update noRag.py
Browse files
noRag.py
CHANGED
@@ -1,120 +1,84 @@
|
|
1 |
-
#
|
2 |
|
3 |
-
import
|
4 |
-
import
|
5 |
from groq import Groq
|
6 |
from pymongo import MongoClient
|
7 |
-
from config import
|
8 |
-
CONNECTION_STRING,
|
9 |
-
CHATGROQ_API_KEY,
|
10 |
-
CUSTOM_PROMPT
|
11 |
-
)
|
12 |
|
13 |
-
#
|
|
|
|
|
|
|
14 |
client = Groq(api_key=CHATGROQ_API_KEY)
|
15 |
mongo = MongoClient(CONNECTION_STRING)
|
16 |
db = mongo["edulearnai"]
|
17 |
chats = db["chats"]
|
18 |
|
|
|
19 |
SYSTEM_PROMPT = "You are a helpful assistant which helps people in their tasks."
|
20 |
|
21 |
-
#
|
22 |
-
|
23 |
-
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
25 |
if not doc:
|
26 |
-
doc = {"session_id": session_id, "history": [], "summary": ""}
|
27 |
chats.insert_one(doc)
|
28 |
-
return doc
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
chats.replace_one({"session_id": doc["session_id"]}, doc)
|
33 |
|
34 |
-
#
|
35 |
-
async def summarize_history(prev_summary: str, history_msgs: list[str]) -> str:
|
36 |
-
"""Ask the LLM to produce a short summary of the combined previous summary + new messages."""
|
37 |
-
combined = prev_summary + "\n" + "\n".join(history_msgs)
|
38 |
-
prompt = (
|
39 |
-
"Summarize the following chat history in one or two short sentences:\n\n"
|
40 |
-
f"{combined}\n\nSummary:"
|
41 |
-
)
|
42 |
-
resp = client.chat.completions.create(
|
43 |
-
model="meta-llama/llama-4-scout-17b-16e-instruct",
|
44 |
-
messages=[{"role": "user", "content": prompt}],
|
45 |
-
temperature=0.3,
|
46 |
-
max_completion_tokens=150,
|
47 |
-
top_p=1,
|
48 |
-
stream=False,
|
49 |
-
)
|
50 |
-
# the first (and only) completion
|
51 |
-
return resp.choices[0].message.content.strip()
|
52 |
-
|
53 |
-
# --- Core chat logic ---
|
54 |
-
async def chat(session_id: str, question: str):
|
55 |
-
session = get_session(session_id)
|
56 |
-
history = session["history"]
|
57 |
-
summary = session["summary"]
|
58 |
-
|
59 |
-
# If history is too long, summarize it and clear
|
60 |
if len(history) >= 10:
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
history = []
|
66 |
|
67 |
-
# Build
|
68 |
-
|
69 |
full_prompt = CUSTOM_PROMPT.format(
|
70 |
context=SYSTEM_PROMPT,
|
71 |
-
chat_history=
|
72 |
-
question=question
|
73 |
)
|
74 |
|
75 |
-
# Call
|
76 |
-
|
77 |
model="meta-llama/llama-4-scout-17b-16e-instruct",
|
78 |
messages=[{"role": "user", "content": full_prompt}],
|
79 |
temperature=1,
|
80 |
max_completion_tokens=1024,
|
81 |
top_p=1,
|
82 |
-
stream=
|
83 |
)
|
|
|
84 |
|
85 |
-
#
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
print() # newline after done
|
93 |
-
|
94 |
-
# Persist the new exchange
|
95 |
-
session["history"].append({"role": "user", "content": question})
|
96 |
-
session["history"].append({"role": "assistant", "content": assistant_response})
|
97 |
-
save_session(session)
|
98 |
-
|
99 |
-
# --- CLI loop ---
|
100 |
-
if __name__ == "__main__":
|
101 |
-
import argparse
|
102 |
-
|
103 |
-
parser = argparse.ArgumentParser()
|
104 |
-
parser.add_argument(
|
105 |
-
"--session",
|
106 |
-
"-s",
|
107 |
-
default="default",
|
108 |
-
help="Session ID (used to key chat history in MongoDB)"
|
109 |
)
|
110 |
-
args = parser.parse_args()
|
111 |
|
112 |
-
|
113 |
-
try:
|
114 |
-
while True:
|
115 |
-
user_q = input("You: ")
|
116 |
-
if not user_q.strip():
|
117 |
-
continue
|
118 |
-
asyncio.run(chat(args.session, user_q))
|
119 |
-
except KeyboardInterrupt:
|
120 |
-
print("\nGoodbye!")
|
|
|
1 |
+
# norag_router.py
|
2 |
|
3 |
+
from fastapi import APIRouter, HTTPException
|
4 |
+
from pydantic import BaseModel
|
5 |
from groq import Groq
|
6 |
from pymongo import MongoClient
|
7 |
+
from config import CONNECTION_STRING, CHATGROQ_API_KEY, CUSTOM_PROMPT
|
|
|
|
|
|
|
|
|
8 |
|
9 |
+
# Create router under /norag
|
10 |
+
router = APIRouter(prefix="/norag", tags=["noRag"])
|
11 |
+
|
12 |
+
# Initialize Groq client and MongoDB
|
13 |
client = Groq(api_key=CHATGROQ_API_KEY)
|
14 |
mongo = MongoClient(CONNECTION_STRING)
|
15 |
db = mongo["edulearnai"]
|
16 |
chats = db["chats"]
|
17 |
|
18 |
+
# System prompt
|
19 |
SYSTEM_PROMPT = "You are a helpful assistant which helps people in their tasks."
|
20 |
|
21 |
+
# Request model
|
22 |
+
type ChatRequest(BaseModel):
|
23 |
+
session_id: str
|
24 |
+
question: str
|
25 |
+
|
26 |
+
@router.post("/chat", summary="Ask a question to the noRag assistant")
|
27 |
+
async def chat_endpoint(req: ChatRequest):
|
28 |
+
# Fetch or create session
|
29 |
+
doc = chats.find_one({"session_id": req.session_id})
|
30 |
if not doc:
|
31 |
+
doc = {"session_id": req.session_id, "history": [], "summary": ""}
|
32 |
chats.insert_one(doc)
|
|
|
33 |
|
34 |
+
history = doc["history"]
|
35 |
+
summary = doc["summary"]
|
|
|
36 |
|
37 |
+
# Summarize if history too long
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
if len(history) >= 10:
|
39 |
+
msgs = [f"{m['role']}: {m['content']}" for m in history]
|
40 |
+
combined = summary + "\n" + "\n".join(msgs)
|
41 |
+
sum_prompt = (
|
42 |
+
"Summarize the following chat history in one or two short sentences:\n\n"
|
43 |
+
+ combined + "\n\nSummary:"
|
44 |
+
)
|
45 |
+
sum_resp = client.chat.completions.create(
|
46 |
+
model="meta-llama/llama-4-scout-17b-16e-instruct",
|
47 |
+
messages=[{"role": "user", "content": sum_prompt}],
|
48 |
+
temperature=0.3,
|
49 |
+
max_completion_tokens=150,
|
50 |
+
top_p=1,
|
51 |
+
stream=False,
|
52 |
+
)
|
53 |
+
summary = sum_resp.choices[0].message.content.strip()
|
54 |
history = []
|
55 |
|
56 |
+
# Build full prompt
|
57 |
+
chat_hist_text = "\n".join([f"{m['role']}: {m['content']}" for m in history])
|
58 |
full_prompt = CUSTOM_PROMPT.format(
|
59 |
context=SYSTEM_PROMPT,
|
60 |
+
chat_history=chat_hist_text or "(no prior messages)",
|
61 |
+
question=req.question
|
62 |
)
|
63 |
|
64 |
+
# Call model
|
65 |
+
resp = client.chat.completions.create(
|
66 |
model="meta-llama/llama-4-scout-17b-16e-instruct",
|
67 |
messages=[{"role": "user", "content": full_prompt}],
|
68 |
temperature=1,
|
69 |
max_completion_tokens=1024,
|
70 |
top_p=1,
|
71 |
+
stream=False,
|
72 |
)
|
73 |
+
answer = resp.choices[0].message.content.strip()
|
74 |
|
75 |
+
# Update session doc
|
76 |
+
history.append({"role": "user", "content": req.question})
|
77 |
+
history.append({"role": "assistant", "content": answer})
|
78 |
+
chats.replace_one(
|
79 |
+
{"session_id": req.session_id},
|
80 |
+
{"session_id": req.session_id, "history": history, "summary": summary},
|
81 |
+
upsert=True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
)
|
|
|
83 |
|
84 |
+
return {"session_id": req.session_id, "answer": answer, "summary": summary}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|