apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[BugFix] rollback max_tokens and min_tokens when continue to infer (#5053)
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details

Browse Source 
* [BugFix] rollback  max_tokens and min_tokens when continue to infer

* [BugFix] rollback  max_tokens and min_tokens when continue to infer

---------

Co-authored-by: liqinrui <liqinrui@baidu.com>
This commit is contained in:
LiqinruiG
2025-11-17 19:03:09 +08:00
committed by GitHub
parent bd28f18785
commit 9bb4337143
3 changed files with 3 additions and 155 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/entrypoints/engine_client.py
+1 -8
View File
@@ -207,14 +207,7 @@ class EngineClient:
task["prompt_token_ids_len"] = len(task["prompt_token_ids"])
input_ids_len = task["prompt_token_ids_len"]
completion_token_len = len(task["completion_token_ids"]) if task.get("completion_token_ids") else 0
task["max_tokens"] = min(
self.max_model_len - input_ids_len, max(0, task.get("max_tokens") - completion_token_len)
)
if task.get("min_tokens") is not None:
task["min_tokens"] = max(1, task["min_tokens"] - completion_token_len)
task["max_tokens"] = min(self.max_model_len - input_ids_len, task.get("max_tokens"))
min_tokens = task.get("min_tokens", 1)
if "messages" in task:
del task["messages"]
fastdeploy/input/ernie4_5_vl_processor/ernie4_5_vl_processor.py
+2 -8
View File
@@ -250,9 +250,7 @@ class Ernie4_5_VLProcessor(Ernie4_5Processor):
else:
raise ValueError(f"Request must contain 'prompt', or 'messages': {request}")
completion_token_len = 0
if request.get("completion_token_ids"):
completion_token_len = len(request.get("completion_token_ids"))
self.append_completion_tokens(outputs, request["completion_token_ids"])
outputs = self.pack_outputs(outputs)
@@ -264,16 +262,12 @@ class Ernie4_5_VLProcessor(Ernie4_5Processor):
if max_model_len is not None and len(request["prompt_token_ids"]) > max_model_len:
request["prompt_token_ids"] = request["prompt_token_ids"][: max_model_len - 1]
tmp_max_tokens = 0
if request.get("max_tokens") is None:
request["max_tokens"] = max(1, max_model_len - len(request["prompt_token_ids"]))
tmp_max_tokens = request["max_tokens"]
else:
tmp_max_tokens = min(
max_model_len - len(request["prompt_token_ids"]), max(0, request["max_tokens"] - completion_token_len)
)
request["max_tokens"] = min(max_model_len - len(request["prompt_token_ids"]), request["max_tokens"])
if request.get("reasoning_max_tokens") is None:
request["reasoning_max_tokens"] = max(int(tmp_max_tokens * 0.8), 1)
request["reasoning_max_tokens"] = max(int(request["max_tokens"] * 0.8), 1)
data_processor_logger.info(f"Processed request {request}")
return request