apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-05-08 16:32:41 +08:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

[Feature] bad words support v1 scheduler and specifiy token ids (#3608)

Browse Source 
* support bad_words_token_ids

* docs

* fix test

* fix

* bad words support kvcache v1 and token ids

* fix
This commit is contained in:
Sunny-bot1
2025-08-26 11:14:51 +08:00
committed by GitHub
parent c43a4bec00
commit c68c3c4b8b
16 changed files with 420 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ci_use/EB_Lite/test_EB_Lite_serving.py
+87 -2
View File
@@ -847,7 +847,24 @@ def test_non_streaming_chat_with_bad_words(openai_client, capsys):
assert hasattr(response_1.choices[0], "message")
assert hasattr(response_1.choices[0].message, "completion_token_ids")
assert isinstance(response_1.choices[0].message.completion_token_ids, list)
response_2 = openai_client.chat.completions.create(
model="default",
messages=[{"role": "user", "content": "Hello, how are you?"}],
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=False,
)
assert hasattr(response_2, "choices")
assert len(response_2.choices) > 0
assert hasattr(response_2.choices[0], "message")
assert hasattr(response_2.choices[0].message, "completion_token_ids")
assert isinstance(response_2.choices[0].message.completion_token_ids, list)
assert not any(ids in response_1.choices[0].message.completion_token_ids for ids in bad_token_ids)
assert not any(ids in response_2.choices[0].message.completion_token_ids for ids in bad_token_ids)
def test_streaming_chat_with_bad_words(openai_client, capsys):
@@ -906,7 +923,34 @@ def test_streaming_chat_with_bad_words(openai_client, capsys):
assert isinstance(chunk.choices[0].delta.completion_token_ids, list)
output_tokens_1.append(chunk.choices[0].delta.content)
output_ids_1.extend(chunk.choices[0].delta.completion_token_ids)
response_2 = openai_client.chat.completions.create(
model="default",
messages=[{"role": "user", "content": "Hello, how are you?"}],
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=True,
)
output_tokens_2 = []
output_ids_2 = []
is_first_chunk = True
for chunk in response_2:
assert hasattr(chunk, "choices")
assert len(chunk.choices) > 0
assert hasattr(chunk.choices[0], "delta")
assert hasattr(chunk.choices[0].delta, "content")
assert hasattr(chunk.choices[0].delta, "completion_token_ids")
if is_first_chunk:
is_first_chunk = False
else:
assert isinstance(chunk.choices[0].delta.completion_token_ids, list)
output_tokens_2.append(chunk.choices[0].delta.content)
output_ids_2.extend(chunk.choices[0].delta.completion_token_ids)
assert not any(ids in output_ids_1 for ids in bad_token_ids)
assert not any(ids in output_ids_2 for ids in bad_token_ids)
def test_non_streaming_completion_with_bad_words(openai_client, capsys):
@@ -956,9 +1000,25 @@ def test_non_streaming_completion_with_bad_words(openai_client, capsys):
)
assert hasattr(response_1, "choices")
assert len(response_1.choices) > 0
assert hasattr(response_0.choices[0], "completion_token_ids")
assert isinstance(response_0.choices[0].completion_token_ids, list)
assert hasattr(response_1.choices[0], "completion_token_ids")
assert isinstance(response_1.choices[0].completion_token_ids, list)
response_2 = openai_client.completions.create(
model="default",
prompt="Hello, how are you?",
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=False,
)
assert hasattr(response_2, "choices")
assert len(response_2.choices) > 0
assert hasattr(response_2.choices[0], "completion_token_ids")
assert isinstance(response_2.choices[0].completion_token_ids, list)
assert not any(ids in response_1.choices[0].completion_token_ids for ids in bad_token_ids)
assert not any(ids in response_2.choices[0].completion_token_ids for ids in bad_token_ids)
def test_streaming_completion_with_bad_words(openai_client, capsys):
@@ -1013,7 +1073,32 @@ def test_streaming_completion_with_bad_words(openai_client, capsys):
assert hasattr(chunk.choices[0], "completion_token_ids")
output_tokens_1.append(chunk.choices[0].text)
output_ids_1.extend(chunk.choices[0].completion_token_ids)
# add bad words token ids
response_2 = openai_client.completions.create(
model="default",
prompt="Hello, how are you?",
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=True,
)
output_tokens_2 = []
output_ids_2 = []
is_first_chunk = True
for chunk in response_2:
if is_first_chunk:
is_first_chunk = False
else:
assert hasattr(chunk, "choices")
assert len(chunk.choices) > 0
assert hasattr(chunk.choices[0], "text")
assert hasattr(chunk.choices[0], "completion_token_ids")
output_tokens_2.append(chunk.choices[0].text)
output_ids_2.extend(chunk.choices[0].completion_token_ids)
assert not any(ids in output_ids_1 for ids in bad_token_ids)
assert not any(ids in output_ids_2 for ids in bad_token_ids)
def test_profile_reset_block_num():
tests/e2e/test_EB_Lite_serving.py
+87 -2
View File
@@ -842,7 +842,24 @@ def test_non_streaming_chat_with_bad_words(openai_client, capsys):
assert hasattr(response_1.choices[0], "message")
assert hasattr(response_1.choices[0].message, "completion_token_ids")
assert isinstance(response_1.choices[0].message.completion_token_ids, list)
response_2 = openai_client.chat.completions.create(
model="default",
messages=[{"role": "user", "content": "Hello, how are you?"}],
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=False,
)
assert hasattr(response_2, "choices")
assert len(response_2.choices) > 0
assert hasattr(response_2.choices[0], "message")
assert hasattr(response_2.choices[0].message, "completion_token_ids")
assert isinstance(response_2.choices[0].message.completion_token_ids, list)
assert not any(ids in response_1.choices[0].message.completion_token_ids for ids in bad_token_ids)
assert not any(ids in response_2.choices[0].message.completion_token_ids for ids in bad_token_ids)
def test_streaming_chat_with_bad_words(openai_client, capsys):
@@ -901,7 +918,34 @@ def test_streaming_chat_with_bad_words(openai_client, capsys):
assert isinstance(chunk.choices[0].delta.completion_token_ids, list)
output_tokens_1.append(chunk.choices[0].delta.content)
output_ids_1.extend(chunk.choices[0].delta.completion_token_ids)
response_2 = openai_client.chat.completions.create(
model="default",
messages=[{"role": "user", "content": "Hello, how are you?"}],
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=True,
)
output_tokens_2 = []
output_ids_2 = []
is_first_chunk = True
for chunk in response_2:
assert hasattr(chunk, "choices")
assert len(chunk.choices) > 0
assert hasattr(chunk.choices[0], "delta")
assert hasattr(chunk.choices[0].delta, "content")
assert hasattr(chunk.choices[0].delta, "completion_token_ids")
if is_first_chunk:
is_first_chunk = False
else:
assert isinstance(chunk.choices[0].delta.completion_token_ids, list)
output_tokens_2.append(chunk.choices[0].delta.content)
output_ids_2.extend(chunk.choices[0].delta.completion_token_ids)
assert not any(ids in output_ids_1 for ids in bad_token_ids)
assert not any(ids in output_ids_2 for ids in bad_token_ids)
def test_non_streaming_completion_with_bad_words(openai_client, capsys):
@@ -951,9 +995,25 @@ def test_non_streaming_completion_with_bad_words(openai_client, capsys):
)
assert hasattr(response_1, "choices")
assert len(response_1.choices) > 0
assert hasattr(response_0.choices[0], "completion_token_ids")
assert isinstance(response_0.choices[0].completion_token_ids, list)
assert hasattr(response_1.choices[0], "completion_token_ids")
assert isinstance(response_1.choices[0].completion_token_ids, list)
response_2 = openai_client.completions.create(
model="default",
prompt="Hello, how are you?",
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=False,
)
assert hasattr(response_2, "choices")
assert len(response_2.choices) > 0
assert hasattr(response_2.choices[0], "completion_token_ids")
assert isinstance(response_2.choices[0].completion_token_ids, list)
assert not any(ids in response_1.choices[0].completion_token_ids for ids in bad_token_ids)
assert not any(ids in response_2.choices[0].completion_token_ids for ids in bad_token_ids)
def test_streaming_completion_with_bad_words(openai_client, capsys):
@@ -1008,7 +1068,32 @@ def test_streaming_completion_with_bad_words(openai_client, capsys):
assert hasattr(chunk.choices[0], "completion_token_ids")
output_tokens_1.append(chunk.choices[0].text)
output_ids_1.extend(chunk.choices[0].completion_token_ids)
# add bad words token ids
response_2 = openai_client.completions.create(
model="default",
prompt="Hello, how are you?",
temperature=1,
top_p=0.0,
max_tokens=20,
extra_body={"bad_words_token_ids": bad_token_ids, "return_token_ids": True},
stream=True,
)
output_tokens_2 = []
output_ids_2 = []
is_first_chunk = True
for chunk in response_2:
if is_first_chunk:
is_first_chunk = False
else:
assert hasattr(chunk, "choices")
assert len(chunk.choices) > 0
assert hasattr(chunk.choices[0], "text")
assert hasattr(chunk.choices[0], "completion_token_ids")
output_tokens_2.append(chunk.choices[0].text)
output_ids_2.extend(chunk.choices[0].completion_token_ids)
assert not any(ids in output_ids_1 for ids in bad_token_ids)
assert not any(ids in output_ids_2 for ids in bad_token_ids)
def test_profile_reset_block_num():