secforge / app /modules /llm.py
rk147's picture
Update app/modules/llm.py
e5e4bf3 verified
Raw History Blame Contribute Delete
3.97 kB
"""LLM interface — text, vision, and chat."""
import json
import base64
import streamlit as st
from groq import Groq
from app.config import GROQ_API_KEY, MODEL_NAME, VISION_MODEL_NAME, MAX_TOKENS, TEMPERATURE
from app.utils.prompts import (
EXPLAIN_CODE_WITH_LANG, CLASSIFY_ISSUE_WITH_LANG, PROPOSE_FIX_WITH_LANG,
build_chat_prompt,
)
@st.cache_resource
def get_client() -> Groq:
return Groq(api_key=GROQ_API_KEY)
def _call(prompt: str, max_tokens: int = MAX_TOKENS) -> str:
client = get_client()
resp = client.chat.completions.create(
model=MODEL_NAME,
messages=[{"role": "user", "content": prompt}],
max_tokens=max_tokens,
temperature=TEMPERATURE,
)
return resp.choices[0].message.content.strip()
def _parse_json(text: str) -> dict:
try:
start = text.find("{")
end = text.rfind("}") + 1
return json.loads(text[start:end])
except Exception:
return {"error": "Parse failed", "raw": text[:200]}
def explain_code(code: str, language: str = "python") -> dict:
prompt = EXPLAIN_CODE_WITH_LANG.format(language=language, code=code)
return _parse_json(_call(prompt))
def classify_issue(issue: dict, code: str, language: str = "python") -> dict:
prompt = CLASSIFY_ISSUE_WITH_LANG.format(
language=language,
issue=issue.get("message", ""),
code=code,
)
return _parse_json(_call(prompt, max_tokens=200))
def propose_fix(issue: dict, code: str, language: str = "python") -> str:
prompt = PROPOSE_FIX_WITH_LANG.format(
language=language,
issue=issue.get("message", ""),
code=code,
)
return _call(prompt, max_tokens=512)
def chat_response(
user_query: str,
chat_history: list,
context: str = "",
linked_files: list = None,
) -> str:
"""
Chat response with optional file context.
Injects a clear system note about available files so model
never claims it cannot see them.
"""
client = get_client()
system, messages = build_chat_prompt(chat_history, user_query, context)
if linked_files:
file_list = ", ".join(linked_files)
system += (
f"\n\n[System: The user has loaded the following files into this session: {file_list}. "
f"Relevant excerpts from these files are provided above in the context block. "
f"You have full access to their contents via that context. "
f"NEVER say you cannot access files or that you are text-based only.]"
)
resp = client.chat.completions.create(
model=MODEL_NAME,
messages=[{"role": "system", "content": system}] + messages,
max_tokens=MAX_TOKENS,
temperature=0.3,
)
return resp.choices[0].message.content.strip()
def vision_response(
user_query: str,
image_bytes: bytes,
media_type: str = "image/png",
context: str = "",
) -> str:
"""Send an image + query to the vision model."""
client = get_client()
b64 = base64.b64encode(image_bytes).decode("utf-8")
system = "You are an expert software engineer and code analyst. Answer questions about code, architecture diagrams, error screenshots, and related technical images clearly and concisely."
if context:
system += f"\n\nAdditional context from linked files:\n```\n{context}\n```"
resp = client.chat.completions.create(
model=VISION_MODEL_NAME,
messages=[
{"role": "system", "content": system},
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {"url": f"data:{media_type};base64,{b64}"},
},
{"type": "text", "text": user_query},
],
},
],
max_tokens=MAX_TOKENS,
temperature=0.3,
)
return resp.choices[0].message.content.strip()