apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-26 10:00:33 +08:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

support reasoning_max_tokens (#4207)

Browse Source
This commit is contained in:
Yuanle Liu
2025-09-23 15:44:41 +08:00
committed by GitHub
parent 0bfffdbc14
commit 361104508e
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/entrypoints/engine_client.py
View File

@@ -149,8 +149,6 @@ class EngineClient:
task["prompt_token_ids_len"] = len(task["prompt_token_ids"]) task["prompt_token_ids_len"] = len(task["prompt_token_ids"])
input_ids_len = task["prompt_token_ids_len"] input_ids_len = task["prompt_token_ids_len"]
task["max_tokens"] = min(self.max_model_len - input_ids_len, task.get("max_tokens")) task["max_tokens"] = min(self.max_model_len - input_ids_len, task.get("max_tokens"))
if task.get("reasoning_max_tokens", None) is None:
task["reasoning_max_tokens"] = max(int(task["max_tokens"] * 0.8), 1)
min_tokens = task.get("min_tokens", 1) min_tokens = task.get("min_tokens", 1)
if "messages" in task: if "messages" in task:
del task["messages"] del task["messages"]
@@ -229,8 +227,8 @@ class EngineClient:
raise ValueError(f"max_tokens can be defined [1, {self.max_model_len}).") raise ValueError(f"max_tokens can be defined [1, {self.max_model_len}).")
if data.get("reasoning_max_tokens") is not None: if data.get("reasoning_max_tokens") is not None:
if data["reasoning_max_tokens"] > data["max_tokens"] or data["reasoning_max_tokens"] < 1: if data["reasoning_max_tokens"] > data["max_tokens"] or data["reasoning_max_tokens"] < 0:
raise ValueError("reasoning_max_tokens must be between max_tokens and 1") raise ValueError("reasoning_max_tokens must be between max_tokens and 0")
if data.get("top_p") is not None: if data.get("top_p") is not None:
if data["top_p"] > 1 or data["top_p"] < 0: if data["top_p"] > 1 or data["top_p"] < 0: