azure openai compatibility

routstr/payment/helpers.py

@@ -1,5 +1,6 @@
 import json
 import os
+from typing import Mapping
 
 from fastapi import HTTPException, Response
 from fastapi.requests import Request
@@ -14,6 +15,7 @@ logger = get_logger(__name__)
 
 UPSTREAM_BASE_URL = os.environ.get("UPSTREAM_BASE_URL", "")
 UPSTREAM_API_KEY = os.environ.get("UPSTREAM_API_KEY", "")
+CHAT_COMPLETIONS_API_VERSION = os.environ.get("CHAT_COMPLETIONS_API_VERSION", "")
 
 if not UPSTREAM_BASE_URL:
     raise ValueError("Please set the UPSTREAM_BASE_URL environment variable")
@@ -201,3 +203,13 @@ def prepare_upstream_headers(request_headers: dict) -> dict:
     )
 
     return headers
+
+
+def prepare_upstream_params(
+    path: str, query_params: Mapping[str, str] | None
+) -> dict[str, str]:
+    """Prepare query params for upstream request, optionally adding api-version for chat/completions."""
+    params: dict[str, str] = dict(query_params or {})
+    if path.endswith("chat/completions") and CHAT_COMPLETIONS_API_VERSION:
+        params["api-version"] = CHAT_COMPLETIONS_API_VERSION
+    return params

routstr/payment/x_cashu.py

@@ -9,7 +9,12 @@ from fastapi.responses import Response, StreamingResponse
 from ..core import get_logger
 from ..wallet import recieve_token, send_token
 from .cost_caculation import CostData, CostDataError, MaxCostData, calculate_cost
-from .helpers import UPSTREAM_BASE_URL, create_error_response, prepare_upstream_headers
+from .helpers import (
+    UPSTREAM_BASE_URL,
+    create_error_response,
+    prepare_upstream_headers,
+    prepare_upstream_params,
+)
 
 logger = get_logger(__name__)
 
@@ -128,7 +133,7 @@ async def forward_to_upstream(
                     url,
                     headers=headers,
                     content=request.stream(),
-                    params=request.query_params,
+                    params=prepare_upstream_params(path, request.query_params),
                 ),
                 stream=True,
             )

routstr/proxy.py

@@ -21,6 +21,7 @@ from .payment.helpers import (
     create_error_response,
     get_max_cost_for_model,
     prepare_upstream_headers,
+    prepare_upstream_params,
 )
 from .payment.x_cashu import x_cashu_handler
 
@@ -283,7 +284,7 @@ async def forward_to_upstream(
                     url,
                     headers=headers,
                     content=request_body,
-                    params=request.query_params,
+                    params=prepare_upstream_params(path, request.query_params),
                 ),
                 stream=True,
             )
@@ -294,7 +295,7 @@ async def forward_to_upstream(
                     url,
                     headers=headers,
                     content=request.stream(),
-                    params=request.query_params,
+                    params=prepare_upstream_params(path, request.query_params),
                 ),
                 stream=True,
             )
@@ -722,7 +723,7 @@ async def forward_get_to_upstream(
                     url,
                     headers=headers,
                     content=request.stream(),
-                    params=request.query_params,
+                    params=prepare_upstream_params(path, request.query_params),
                 ),
             )