fm_proxy

proxy.py

from http.server import BaseHTTPRequestHandler, HTTPServer
from http.client import HTTPConnection
import time
import json

UPSTREAM_HOST = "127.0.0.1"
UPSTREAM_PORT = 1976

class Proxy(BaseHTTPRequestHandler):
    def do_GET(self):
        path = self.path
        if path == "/v1/models?":
            path = "/v1/models"

        conn = HTTPConnection(UPSTREAM_HOST, UPSTREAM_PORT)
        conn.request("GET", path, headers=self.headers)
        resp = conn.getresponse()

        self.send_response(resp.status)
        for k, v in resp.getheaders():
            if k.lower() != "transfer-encoding":
                self.send_header(k, v)
        self.end_headers()
        self.wfile.write(resp.read())
        
    def do_POST(self):
        length = int(self.headers.get("Content-Length", 0))
        body = self.rfile.read(length)

        path = self.path

        conn = HTTPConnection(UPSTREAM_HOST, UPSTREAM_PORT)
        conn.request("POST", path, body=body, headers=dict(self.headers))

        resp = conn.getresponse()

        self.send_response(resp.status)

        for k, v in resp.getheaders():
            if k.lower() == "transfer-encoding":
                continue
            self.send_header(k, v)

        self.end_headers()

        while True:
            chunk = resp.read(4096)
            if not chunk:
                break

            text = chunk.decode("utf-8", errors="ignore")

            for line in text.splitlines():
                if not line.startswith("data:"):
                    continue

                payload = line[6:].strip()

                if payload == "[DONE]":
                    continue

                obj = json.loads(payload)

                # 🔥 PUT IT HERE
                obj.setdefault("object", "chat.completion.chunk")
                obj.setdefault("created", int(time.time()))

                choice = obj["choices"][0]
                choice.setdefault("index", 0)
                choice.setdefault("finish_reason", None)

                delta = choice.setdefault("delta", {})
                delta.setdefault("content", None)

                # re-emit SSE
                self.wfile.write(f"data: {json.dumps(obj)}\n\n".encode())
                self.wfile.flush()

HTTPServer(("127.0.0.1", 1977), Proxy).serve_forever()