apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 00:57:33 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[BugFix] fix parameter is 0 (#3663)

Browse Source 
* Update engine.py

* Update engine_client.py
This commit is contained in:
ltd0924
2025-08-28 09:52:17 +08:00
committed by GitHub
parent 47e6270dec
commit 6a90cfd144
2 changed files with 31 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
fastdeploy/engine/engine.py
View File

@@ -400,10 +400,10 @@ class LLMEngine:
while self.running:
try:
block = True if len(added_requests) == 0 else False
if not self.cfg.enable_mm:
err, data = self.recv_request_server.receive_json_once(block)
if not self.cfg.model_config.enable_mm:
err, data = self.zmq_server.receive_json_once(block)
else:
err, data = self.recv_request_server.receive_pyobj_once(block)
err, data = self.zmq_server.receive_pyobj_once(block)
if err is not None:
llm_logger.error("Engine stops inserting zmq task into scheduler, err:{err}")
break
@@ -411,19 +411,23 @@ class LLMEngine:
request, insert_task = None, []
results: List[Tuple[str, Optional[str]]] = list()
if data:
request = Request.from_dict(data)
start_span("ENQUEUE_ZMQ", data, trace.SpanKind.PRODUCER)
llm_logger.debug(f"Receive request: {request}")
err_msg = None
if self.guided_decoding_checker is not None:
request, err_msg = self.guided_decoding_checker.schema_format(request)
try:
request = Request.from_dict(data)
start_span("ENQUEUE_ZMQ", data, trace.SpanKind.PRODUCER)
llm_logger.debug(f"Receive request: {request}")
except Exception as e:
llm_logger.error(f"Receive request error: {e}, {traceback.format_exc()!s}")
err_msg = str(e)
results.append((data["request_id"], err_msg))
if err_msg is not None:
llm_logger.error(err_msg)
results.append((request.request_id, err_msg))
else:
if self.guided_decoding_checker is not None and err_msg is None:
request, err_msg = self.guided_decoding_checker.schema_format(request)
if err_msg is not None:
llm_logger.error(f"Receive request error: {err_msg}")
results.append((request.request_id, err_msg))
if err_msg is None:
insert_task.append(request)
response = self.scheduler.put_requests(insert_task)
@@ -435,9 +439,10 @@ class LLMEngine:
added_requests[request.request_id] += 1
for request_id, failed in results:
added_requests[request_id] -= 1
if added_requests[request_id] == 0:
added_requests.pop(request_id)
if request_id in added_requests:
added_requests[request_id] -= 1
if added_requests[request_id] == 0:
added_requests.pop(request_id)
if failed is None:
main_process_metrics.num_requests_waiting.inc(1)
@@ -451,7 +456,7 @@ class LLMEngine:
)
# Since the request is not in scheduler
# Send result by zmq directly
self.send_response_server.send_response(request_id, [error_result])
self.zmq_server.send_multipart(request_id, [error_result])
except Exception as e:
llm_logger.error(
f"Error happend while receving new request from zmq, details={e}, "

16
fastdeploy/entrypoints/engine_client.py
View File

@@ -190,35 +190,35 @@ class EngineClient:
Validate stream options
"""
if data.get("n"):
if data.get("n") is not None:
if data["n"] != 1:
raise ValueError("n only support 1.")
if data.get("max_tokens"):
if data.get("max_tokens") is not None:
if data["max_tokens"] < 1 or data["max_tokens"] >= self.max_model_len:
raise ValueError(f"max_tokens can be defined [1, {self.max_model_len}).")
if data.get("reasoning_max_tokens"):
if data.get("reasoning_max_tokens") is not None:
if data["reasoning_max_tokens"] > data["max_tokens"] or data["reasoning_max_tokens"] < 1:
raise ValueError("reasoning_max_tokens must be between max_tokens and 1")
if data.get("top_p"):
if data.get("top_p") is not None:
if data["top_p"] > 1 or data["top_p"] < 0:
raise ValueError("top_p value can only be defined [0, 1].")
if data.get("frequency_penalty"):
if data.get("frequency_penalty") is not None:
if not -2.0 <= data["frequency_penalty"] <= 2.0:
raise ValueError("frequency_penalty must be in [-2, 2]")
if data.get("temperature"):
if data.get("temperature") is not None:
if data["temperature"] < 0:
raise ValueError("temperature must be non-negative")
if data.get("presence_penalty"):
if data.get("presence_penalty") is not None:
if not -2.0 <= data["presence_penalty"] <= 2.0:
raise ValueError("presence_penalty must be in [-2, 2]")
if data.get("seed"):
if data.get("seed") is not None:
if not 0 <= data["seed"] <= 922337203685477580:
raise ValueError("seed must be in [0, 922337203685477580]")