Mercurial
annotate mrjunejune/inference/litellm_proxy.py @ 260:1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
Integrate the production JRPG chat with Seobeo streaming, Deita persistence, and a Bazel-managed Copilot SDK and LiteLLM inference stack.
Co-authored-by: Copilot <[email protected]>
| author | MrJuneJune <mrjunejune@users.noreply.github.com> |
|---|---|
| date | Wed, 05 Aug 2026 09:19:41 -0700 |
| parents | |
| children |
| rev | line source |
|---|---|
|
260
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
1 from __future__ import annotations |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
2 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
3 import argparse |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
4 import asyncio |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
5 import json |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
6 import os |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
7 import pathlib |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
8 from collections.abc import AsyncIterator, Awaitable, Callable |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
9 from typing import Any |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
10 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
11 import litellm |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
12 import uvicorn |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
13 import yaml |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
14 from fastapi import FastAPI, HTTPException, Request |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
15 from fastapi.responses import JSONResponse, StreamingResponse |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
16 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
17 Completion = Callable[..., Awaitable[Any]] |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
18 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
19 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
20 def resolve_token_directory(explicit: str | None = None) -> pathlib.Path: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
21 if explicit: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
22 return pathlib.Path(explicit).expanduser().resolve() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
23 configured = os.environ.get("GITHUB_COPILOT_TOKEN_DIR") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
24 if configured: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
25 return pathlib.Path(configured).expanduser().resolve() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
26 state_root = os.environ.get("MRJUNEJUNE_INFERENCE_STATE") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
27 if state_root: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
28 return ( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
29 pathlib.Path(state_root).expanduser().resolve() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
30 / "litellm-copilot" |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
31 ) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
32 xdg_state = os.environ.get("XDG_STATE_HOME") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
33 root = ( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
34 pathlib.Path(xdg_state).expanduser() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
35 if xdg_state |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
36 else pathlib.Path.home() / ".local" / "state" |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
37 ) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
38 return (root / "mrjunejune" / "inference" / "litellm-copilot").resolve() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
39 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
40 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
41 def _serialize(value: Any) -> dict[str, Any]: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
42 if hasattr(value, "model_dump"): |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
43 return value.model_dump(exclude_none=True) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
44 if hasattr(value, "dict"): |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
45 return value.dict() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
46 if isinstance(value, dict): |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
47 return value |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
48 raise TypeError("LiteLLM returned an unsupported response") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
49 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
50 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
51 def create_app( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
52 completion: Completion = litellm.acompletion, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
53 *, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
54 model_alias: str = "jrpg-copilot", |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
55 upstream_model: str = "github_copilot/gpt-4", |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
56 ) -> FastAPI: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
57 app = FastAPI(docs_url=None, redoc_url=None, openapi_url=None) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
58 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
59 @app.get("/health/liveliness") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
60 async def health() -> dict[str, str]: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
61 return {"status": "ready"} |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
62 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
63 @app.get("/health/readiness") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
64 async def readiness() -> dict[str, str]: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
65 from litellm.llms.github_copilot.authenticator import Authenticator |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
66 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
67 try: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
68 await asyncio.to_thread(Authenticator().get_api_key) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
69 except Exception as error: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
70 raise HTTPException( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
71 status_code=503, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
72 detail="GitHub Copilot authentication unavailable", |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
73 ) from error |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
74 return {"status": "ready"} |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
75 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
76 @app.post("/v1/chat/completions") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
77 async def chat(request: Request): |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
78 master_key = os.environ.get("LITELLM_MASTER_KEY", "") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
79 authorization = request.headers.get("Authorization", "") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
80 if not master_key or authorization != f"Bearer {master_key}": |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
81 raise HTTPException(status_code=401, detail="Invalid gateway key") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
82 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
83 payload = await request.json() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
84 if not isinstance(payload, dict) or payload.get("model") != model_alias: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
85 raise HTTPException(status_code=400, detail="Unknown model alias") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
86 payload = dict(payload) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
87 payload["model"] = upstream_model |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
88 configured_max = int(os.environ.get("LITELLM_MAX_OUTPUT_TOKENS", "1024")) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
89 requested_max = payload.get("max_tokens") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
90 if not isinstance(requested_max, int) or requested_max > configured_max: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
91 payload["max_tokens"] = configured_max |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
92 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
93 try: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
94 response = await completion(**payload) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
95 except Exception as error: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
96 raise HTTPException( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
97 status_code=502, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
98 detail="GitHub Copilot provider request failed", |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
99 ) from error |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
100 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
101 if payload.get("stream"): |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
102 async def events() -> AsyncIterator[str]: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
103 async for chunk in response: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
104 yield ( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
105 "data: " |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
106 + json.dumps(_serialize(chunk), separators=(",", ":")) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
107 + "\n\n" |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
108 ) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
109 yield "data: [DONE]\n\n" |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
110 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
111 return StreamingResponse(events(), media_type="text/event-stream") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
112 return JSONResponse(_serialize(response)) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
113 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
114 return app |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
115 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
116 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
117 def _load_model(config_path: str) -> tuple[str, str]: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
118 with open(config_path, encoding="utf-8") as config_file: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
119 config = yaml.safe_load(config_file) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
120 models = config.get("model_list", []) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
121 if len(models) != 1: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
122 raise ValueError("LiteLLM config must contain exactly one model") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
123 model = models[0] |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
124 return model["model_name"], model["litellm_params"]["model"] |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
125 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
126 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
127 def main() -> None: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
128 parser = argparse.ArgumentParser(description="Minimal LiteLLM gateway") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
129 parser.add_argument("--host", default="127.0.0.1") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
130 parser.add_argument("--port", type=int, default=4000) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
131 parser.add_argument("--config") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
132 parser.add_argument("--authenticate", action="store_true") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
133 parser.add_argument( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
134 "--token-dir", |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
135 help=( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
136 "Persistent GitHub Copilot token directory. Defaults to " |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
137 "GITHUB_COPILOT_TOKEN_DIR, MRJUNEJUNE_INFERENCE_STATE, or " |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
138 "~/.local/state/mrjunejune/inference/litellm-copilot." |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
139 ), |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
140 ) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
141 args = parser.parse_args() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
142 if args.authenticate: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
143 from litellm.llms.github_copilot.authenticator import Authenticator |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
144 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
145 token_directory = resolve_token_directory(args.token_dir) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
146 token_directory.mkdir(parents=True, exist_ok=True, mode=0o700) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
147 token_directory.chmod(0o700) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
148 os.environ["GITHUB_COPILOT_TOKEN_DIR"] = str(token_directory) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
149 authenticator = Authenticator() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
150 authenticator.get_access_token() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
151 authenticator.get_api_key() |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
152 print(f"GitHub Copilot authentication stored in {token_directory}") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
153 return |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
154 if not args.config: |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
155 parser.error("--config is required unless --authenticate is used") |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
156 model_alias, upstream_model = _load_model(args.config) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
157 uvicorn.run( |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
158 create_app(model_alias=model_alias, upstream_model=upstream_model), |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
159 host=args.host, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
160 port=args.port, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
161 access_log=False, |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
162 ) |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
163 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
164 |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
165 if __name__ == "__main__": |
|
1f9877b637e9
Add Copilot-powered cyberpunk JRPG chat
MrJuneJune <mrjunejune@users.noreply.github.com>
parents:
diff
changeset
|
166 main() |