apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-13 12:23:55 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

[CP] CP Lm head fp32 and temp_logprob to release/2.1 (#3766)

Browse Source 
* [Feature] Add temp_scaled_logprobs and top_p_normalized_logprobs parameters for logits and logprobs post processing (#3552)

* [feature] Add temp_scaled_logprobs and top_p_normalized_logprobs parameters for logits and logprobs post processing

* infer engine support temp_scaled_logprobs and top_p_normalized_logprobs

* delete some code

* code check

* code check and add doc

* fix tokenizer.decoder(-1), return 'Invalid Token'

* add ci for temp_scaled and top_p logprobs

* check test

* check seq len time shape

* logprob clip inf

---------

Co-authored-by: sunlei1024 <sunlei5788@gmail.com>

* [Precision] Support lm_head layer running in float32 (#3597)

* support lm_head fp32 bf16 fp16

* support lm_head fp32 bf16 fp16

* add doc and check code

* lm_head_fp32 specify lm_head as fp32

* code check

* check doc

* code check

---------

Co-authored-by: sunlei1024 <sunlei5788@gmail.com>
This commit is contained in:
chen
2025-09-01 19:56:54 +08:00
committed by GitHub
parent 4da603daec
commit 1e19833ba5
22 changed files with 188 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
fastdeploy/worker/gpu_model_runner.py
View File

@@ -315,6 +315,10 @@ class GPUModelRunner(ModelRunnerBase):
self.share_inputs["penalty_score"][idx : idx + 1] = request.get("repetition_penalty", 1.0)
self.share_inputs["frequency_score"][idx : idx + 1] = request.get("frequency_penalty", 0.0)
self.share_inputs["presence_score"][idx : idx + 1] = request.get("presence_penalty", 0.0)
self.share_inputs["temp_scaled_logprobs"][idx : idx + 1] = request.get("temp_scaled_logprobs", False)
self.share_inputs["top_p_normalized_logprobs"][idx : idx + 1] = request.get(
"top_p_normalized_logprobs", False
)
self.share_inputs["min_dec_len"][idx : idx + 1] = request.get("min_tokens", 1)
self.share_inputs["max_dec_len"][idx : idx + 1] = request.get(
@@ -493,6 +497,12 @@ class GPUModelRunner(ModelRunnerBase):
self.share_inputs["presence_score"][idx : idx + 1] = get_attr_from_request(
request, "presence_penalty", 0.0
)
self.share_inputs["temp_scaled_logprobs"][idx : idx + 1] = get_attr_from_request(
request, "temp_scaled_logprobs", False
)
self.share_inputs["top_p_normalized_logprobs"][idx : idx + 1] = get_attr_from_request(
request, "top_p_normalized_logprobs", False
)
self.share_inputs["min_dec_len"][idx : idx + 1] = request.get("min_tokens", 1)
self.share_inputs["max_dec_len"][idx : idx + 1] = request.get(
@@ -622,6 +632,8 @@ class GPUModelRunner(ModelRunnerBase):
self.share_inputs["presence_score"] = paddle.full(
[max_num_seqs, 1], self.model_config.presence_score, dtype="float32"
)
self.share_inputs["temp_scaled_logprobs"] = paddle.full([max_num_seqs, 1], False, dtype="bool")
self.share_inputs["top_p_normalized_logprobs"] = paddle.full([max_num_seqs, 1], False, dtype="bool")
self.share_inputs["min_dec_len"] = paddle.full([max_num_seqs, 1], self.model_config.min_length, dtype="int64")
self.share_inputs["max_dec_len"] = paddle.full(
@@ -841,6 +853,9 @@ class GPUModelRunner(ModelRunnerBase):
max_num_logprobs=20 if self.enable_logprob else None,
enable_early_stop=self.enable_early_stop,
stop_flags=self.share_inputs["stop_flags"],
temp_scaled_logprobs=self.share_inputs["temp_scaled_logprobs"],
top_p_normalized_logprobs=self.share_inputs["top_p_normalized_logprobs"],
share_inputs=self.share_inputs,
)
def load_model(self) -> None: