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

[FIX 2.0.2]fix topp topk default value (#2810)

Browse Source 
* fix topp topk default value

* update topk
This commit is contained in:
Sunny-bot1
2025-07-11 16:12:02 +08:00
committed by GitHub
parent ec986642df
commit 4596dd7248
3 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
fastdeploy/engine/sampling_params.py
View File

@@ -82,7 +82,7 @@ class SamplingParams:
frequency_penalty: float = None frequency_penalty: float = None
repetition_penalty: float = None repetition_penalty: float = None
temperature: float = None temperature: float = None
top_p: float = 1.0 top_p: float = None
top_k: int = 0 top_k: int = 0
seed: Optional[int] = None seed: Optional[int] = None
stop: Optional[Union[str, List[str]]] = None stop: Optional[Union[str, List[str]]] = None
@@ -132,7 +132,7 @@ class SamplingParams:
repetition_penalty=repetition_penalty repetition_penalty=repetition_penalty
if repetition_penalty is not None else 1.0, if repetition_penalty is not None else 1.0,
temperature=temperature if temperature is not None else 1.0, temperature=temperature if temperature is not None else 1.0,
top_p=top_p if top_p is not None else 1.0, top_p=top_p,
top_k=top_k if top_k is not None else 0, top_k=top_k if top_k is not None else 0,
seed=seed, seed=seed,
stop=stop, stop=stop,

2
fastdeploy/worker/gpu_model_runner.py
View File

@@ -244,7 +244,7 @@ class GPUModelRunner(ModelRunnerBase):
request.eos_token_ids.append(request.eos_token_ids[0]) request.eos_token_ids.append(request.eos_token_ids[0])
self.share_inputs["eos_token_id"][:] = np.array( self.share_inputs["eos_token_id"][:] = np.array(
request.eos_token_ids, dtype="int64").reshape(-1, 1) request.eos_token_ids, dtype="int64").reshape(-1, 1)
self.share_inputs["top_p"][idx:idx + 1] = request.get("top_p", 1.0) self.share_inputs["top_p"][idx:idx + 1] = request.get("top_p", 0.7)
self.share_inputs["top_k"][idx:idx + 1] = request.get("top_k", 0) self.share_inputs["top_k"][idx:idx + 1] = request.get("top_k", 0)
self.share_inputs["temperature"][idx:idx + 1] = request.get( self.share_inputs["temperature"][idx:idx + 1] = request.get(
"temperature", 0.95) "temperature", 0.95)

2
fastdeploy/worker/xpu_model_runner.py
View File

@@ -279,7 +279,7 @@ class XPUModelRunner(ModelRunnerBase):
self.share_inputs["eos_token_id"][:] = np.array( self.share_inputs["eos_token_id"][:] = np.array(
request.eos_token_ids, dtype="int64").reshape(-1, 1) request.eos_token_ids, dtype="int64").reshape(-1, 1)
self.share_inputs["pre_ids"][idx:idx + 1] = -1 self.share_inputs["pre_ids"][idx:idx + 1] = -1
self.share_inputs["top_p"][idx:idx + 1] = request.get("top_p", 1.0) self.share_inputs["top_p"][idx:idx + 1] = request.get("top_p", 0.7)
self.share_inputs["top_k"][idx:idx + 1] = request.get("top_k", 0) self.share_inputs["top_k"][idx:idx + 1] = request.get("top_k", 0)
self.share_inputs["temperature"][idx:idx + 1] = request.get( self.share_inputs["temperature"][idx:idx + 1] = request.get(
"temperature", 0.95) "temperature", 0.95)