diff --git a/scripts/cache_aware_proxy.py b/scripts/cache_aware_proxy.py
index 629ce49..4a49072 100644
--- a/scripts/cache_aware_proxy.py
+++ b/scripts/cache_aware_proxy.py
@@ -426,16 +426,16 @@ async def _handle_combined(api, req_data, token_ids, input_length, session_id, h
             try:
                 for attempt in range(MAX_STREAM_RETRIES):
                     try:
-                        async with inst.client.stream("POST", api, json=req_data, headers=headers) as resp:
+                        async with chosen.client.stream("POST", api, json=req_data, headers=headers) as resp:
                             resp.raise_for_status()
                             async for chunk in resp.aiter_bytes():
                                 if not prefill_done:
-                                    inst.pending_prefill_tokens -= estimated_new
-                                    inst.ongoing_decode_tokens += input_length
+                                    chosen.pending_prefill_tokens -= estimated_new
+                                    chosen.ongoing_decode_tokens += input_length
                                     breakdown["t_first_token"] = _time.monotonic()
                                     prefill_done = True
                                 yield chunk
-                        inst.record_prefix(token_ids)
+                        chosen.record_prefix(token_ids)
                         break
                     except (httpx.ConnectError, httpx.RemoteProtocolError):
                         if prefill_done or attempt >= MAX_STREAM_RETRIES - 1:
@@ -443,11 +443,11 @@ async def _handle_combined(api, req_data, token_ids, input_length, session_id, h
                         await asyncio.sleep(RETRY_DELAY_S)
             finally:
                 if not prefill_done:
-                    inst.pending_prefill_tokens -= estimated_new
+                    chosen.pending_prefill_tokens -= estimated_new
                 else:
-                    inst.ongoing_decode_tokens -= input_length
-                inst.ongoing_tokens -= input_length
-                inst.num_requests -= 1
+                    chosen.ongoing_decode_tokens -= input_length
+                chosen.ongoing_tokens -= input_length
+                chosen.num_requests -= 1
                 breakdown["t_done"] = _time.monotonic()
                 _breakdown_log.append(breakdown)