apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[FIX]fix bad_words when sending requests consecutively (#3197)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* fix bad_words

* fix log

* fix log
This commit is contained in:
Sunny-bot1
2025-08-04 20:59:41 +08:00
committed by GitHub
parent 1f8289e106
commit 72ef5a9c93
5 changed files with 45 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/engine/sampling_params.py
View File

@@ -218,16 +218,18 @@ class SamplingParams:
prompt_token_ids = tokenizer.encode(text=prompt, add_special_tokens=False)["input_ids"]
if len(prompt_token_ids) != 1:
if not add_prefix_space:
logger.warning(
f"Skip bad_words: {prompt}."
f"Skip bad_words: <{prompt}>."
f"Bad words should be a single token."
f"Got tokens: {prompt_token_ids}."
)
continue
if prompt_token_ids[0] > tokenizer.vocab_size:
if not add_prefix_space:
logger.warning(
f"Skip bad_words: {prompt}."
f"Skip bad_words: <{prompt}>."
f"All token id values should be satisfying:"
f" 0 <= token_id < {tokenizer.vocab_size}."
f"Got token: {prompt_token_ids}."

6
fastdeploy/worker/gcu_model_runner.py
View File

@@ -270,13 +270,15 @@ class GCUModelRunner(ModelRunnerBase):
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"))
if bad_words_len > 0:
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(
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:
stop_seqs_num = len(request.get("stop_seqs_len"))

6
fastdeploy/worker/gpu_model_runner.py
View File

@@ -489,13 +489,15 @@ class GPUModelRunner(ModelRunnerBase):
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"))
if bad_words_len > 0:
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(
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:
stop_seqs_num = len(request.get("stop_seqs_len"))

6
fastdeploy/worker/iluvatar_model_runner.py
View File

@@ -242,13 +242,15 @@ class IluvatarModelRunner(ModelRunnerBase):
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"))
if bad_words_len > 0:
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(
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:
stop_seqs_num = len(request.get("stop_seqs_len"))

6
fastdeploy/worker/xpu_model_runner.py
View File

@@ -506,13 +506,15 @@ class XPUModelRunner(ModelRunnerBase):
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"))
if bad_words_len > 0:
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(
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:
stop_seqs_num = len(request.get("stop_seqs_len"))