apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[FIX 2.1]fix bad_words when sending requests consecutively (#3199)

Browse Source 
* fix bad_words

* fix log

* fix log
This commit is contained in:
Sunny-bot1
2025-08-06 15:47:27 +08:00
committed by GitHub
parent bc0b92bba4
commit f672a34f95
5 changed files with 45 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
fastdeploy/engine/sampling_params.py
View File

@@ -218,20 +218,22 @@ class SamplingParams:
prompt_token_ids = tokenizer.encode(text=prompt, add_special_tokens=False)["input_ids"] prompt_token_ids = tokenizer.encode(text=prompt, add_special_tokens=False)["input_ids"]
if len(prompt_token_ids) != 1: if len(prompt_token_ids) != 1:
logger.warning( if not add_prefix_space:
f"Skip bad_words: {prompt}." logger.warning(
f"Bad words should be a single token." f"Skip bad_words: <{prompt}>."
f"Got tokens: {prompt_token_ids}." f"Bad words should be a single token."
) f"Got tokens: {prompt_token_ids}."
)
continue continue
if prompt_token_ids[0] > tokenizer.vocab_size: if prompt_token_ids[0] > tokenizer.vocab_size:
logger.warning( if not add_prefix_space:
f"Skip bad_words: {prompt}." logger.warning(
f"All token id values should be satisfying:" f"Skip bad_words: <{prompt}>."
f" 0 <= token_id < {tokenizer.vocab_size}." f"All token id values should be satisfying:"
f"Got token: {prompt_token_ids}." f" 0 <= token_id < {tokenizer.vocab_size}."
) f"Got token: {prompt_token_ids}."
)
continue continue
if prompt_token_ids not in self._bad_words_token_ids: if prompt_token_ids not in self._bad_words_token_ids:

14
fastdeploy/worker/gcu_model_runner.py
View File

@@ -272,13 +272,15 @@ class GCUModelRunner(ModelRunnerBase):
request.block_tables, dtype="int32" request.block_tables, dtype="int32"
) )
if request.get("bad_words_token_ids") is not None: if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
bad_words_len = len(request.get("bad_words_token_ids")) bad_words_len = len(request.get("bad_words_token_ids"))
if bad_words_len > 0: self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array( request.get("bad_words_token_ids"), dtype="int64"
request.get("bad_words_token_ids"), dtype="int64" )
) else:
self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None: if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
stop_seqs_num = len(request.get("stop_seqs_len")) stop_seqs_num = len(request.get("stop_seqs_len"))

14
fastdeploy/worker/gpu_model_runner.py
View File

@@ -495,13 +495,15 @@ class GPUModelRunner(ModelRunnerBase):
request.block_tables, dtype="int32" request.block_tables, dtype="int32"
) )
if request.get("bad_words_token_ids") is not None: if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
bad_words_len = len(request.get("bad_words_token_ids")) bad_words_len = len(request.get("bad_words_token_ids"))
if bad_words_len > 0: self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array( request.get("bad_words_token_ids"), dtype="int64"
request.get("bad_words_token_ids"), dtype="int64" )
) else:
self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None: if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
stop_seqs_num = len(request.get("stop_seqs_len")) stop_seqs_num = len(request.get("stop_seqs_len"))

14
fastdeploy/worker/iluvatar_model_runner.py
View File

@@ -243,13 +243,15 @@ class IluvatarModelRunner(ModelRunnerBase):
request.block_tables, dtype="int32" request.block_tables, dtype="int32"
) )
if request.get("bad_words_token_ids") is not None: if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
bad_words_len = len(request.get("bad_words_token_ids")) bad_words_len = len(request.get("bad_words_token_ids"))
if bad_words_len > 0: self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array( request.get("bad_words_token_ids"), dtype="int64"
request.get("bad_words_token_ids"), dtype="int64" )
) else:
self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None: if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
stop_seqs_num = len(request.get("stop_seqs_len")) stop_seqs_num = len(request.get("stop_seqs_len"))

14
fastdeploy/worker/xpu_model_runner.py
View File

@@ -507,13 +507,15 @@ class XPUModelRunner(ModelRunnerBase):
request.block_tables, dtype="int32" request.block_tables, dtype="int32"
) )
if request.get("bad_words_token_ids") is not None: if request.get("bad_words_token_ids") is not None and len(request.get("bad_words_token_ids")) > 0:
bad_words_len = len(request.get("bad_words_token_ids")) bad_words_len = len(request.get("bad_words_token_ids"))
if bad_words_len > 0: self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len
self.share_inputs["bad_tokens_len"][idx : idx + 1] = bad_words_len self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array(
self.share_inputs["bad_tokens"][idx : idx + 1, :bad_words_len] = np.array( request.get("bad_words_token_ids"), dtype="int64"
request.get("bad_words_token_ids"), dtype="int64" )
) else:
self.share_inputs["bad_tokens_len"][idx : idx + 1] = 1
self.share_inputs["bad_tokens"][idx : idx + 1, :] = np.array([-1], dtype="int64")
if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None: if request.get("stop_token_ids") is not None and request.get("stop_seqs_len") is not None:
stop_seqs_num = len(request.get("stop_seqs_len")) stop_seqs_num = len(request.get("stop_seqs_len"))