biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
tests/reasoning/test_mistral_reasoning_parser.py
View File

@@ -3,8 +3,7 @@
import pytest
from mistral_common.tokens.tokenizers.base import SpecialTokens
from mistral_common.tokens.tokenizers.tekken import (SpecialTokenInfo,
Tekkenizer)
from mistral_common.tokens.tokenizers.tekken import SpecialTokenInfo, Tekkenizer
from tests.reasoning.utils import run_reasoning_extraction_mistral
from vllm.reasoning import ReasoningParser, ReasoningParserManager
@@ -18,23 +17,27 @@ def mistral_tokenizer():
# TODO(Julien): upon model release change to a tokenizer already configured.
# =================================================================
mistral_tokenizer = MistralTokenizer.from_pretrained(
"mistralai/Devstral-Small-2507")
"mistralai/Devstral-Small-2507"
)
assert isinstance(mistral_tokenizer.tokenizer, Tekkenizer)
# Add think special tokens to the tokenizer
mistral_tokenizer.tokenizer._all_special_tokens[35] = SpecialTokenInfo(
rank=35, is_control=True, token_str=SpecialTokens.begin_think.value)
rank=35, is_control=True, token_str=SpecialTokens.begin_think.value
)
mistral_tokenizer.tokenizer._all_special_tokens[36] = SpecialTokenInfo(
rank=36, is_control=True, token_str=SpecialTokens.end_think.value)
rank=36, is_control=True, token_str=SpecialTokens.end_think.value
)
mistral_tokenizer.tokenizer._special_tokens_reverse_vocab = {
k: v
for k, v in
mistral_tokenizer.tokenizer._special_tokens_reverse_vocab.items()
for k, v in mistral_tokenizer.tokenizer._special_tokens_reverse_vocab.items()
if v not in {35, 36}
}
mistral_tokenizer.tokenizer._special_tokens_reverse_vocab[
SpecialTokens.begin_think.value] = 35
SpecialTokens.begin_think.value
] = 35
mistral_tokenizer.tokenizer._special_tokens_reverse_vocab[
SpecialTokens.end_think.value] = 36
SpecialTokens.end_think.value
] = 36
mistral_tokenizer.instruct.BEGIN_THINK = 35
mistral_tokenizer.instruct.END_THINK = 36
# =================================================================
@@ -290,39 +293,45 @@ def test_mistral_reasoning(
if index_think != -1:
output_before_think = output[:index_think]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_before_think, False, False)
output_before_think, False, False
)
output_tokens += [mistral_tokenizer.instruct.BEGIN_THINK]
if index_end_think != -1:
output_middle = output[index_think + len_think:index_end_think]
output_after_think = output[index_end_think + len_end_think:]
output_middle = output[index_think + len_think : index_end_think]
output_after_think = output[index_end_think + len_end_think :]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_middle, False, False)
output_middle, False, False
)
output_tokens += [mistral_tokenizer.instruct.END_THINK]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_after_think, False, False)
output_after_think, False, False
)
else:
output_middle = output[index_think + len_think:]
output_middle = output[index_think + len_think :]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_middle, False, False)
output_middle, False, False
)
elif index_end_think != -1:
output_before_think = output[:index_end_think]
output_after_think = output[index_end_think + len_end_think:]
output_after_think = output[index_end_think + len_end_think :]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_before_think, False, False)
output_before_think, False, False
)
output_tokens += [mistral_tokenizer.instruct.END_THINK]
output_tokens += mistral_tokenizer.tokenizer.encode(
output_after_think, False, False)
output_after_think, False, False
)
else:
output_tokens += mistral_tokenizer.tokenizer.encode(
output, False, False)
output_tokens += mistral_tokenizer.tokenizer.encode(output, False, False)
parser: ReasoningParser = ReasoningParserManager.get_reasoning_parser(
parser_name)(mistral_tokenizer)
parser: ReasoningParser = ReasoningParserManager.get_reasoning_parser(parser_name)(
mistral_tokenizer
)
reasoning, content = run_reasoning_extraction_mistral(parser,
output_tokens,
streaming=streaming)
reasoning, content = run_reasoning_extraction_mistral(
parser, output_tokens, streaming=streaming
)
assert reasoning == param_dict["reasoning_content"]
assert content == param_dict["content"]
@@ -335,7 +344,8 @@ def test_mistral_reasoning(
if param_dict["content"] is not None:
content = parser.extract_content_ids(output_tokens)
assert content == mistral_tokenizer.tokenizer.encode(
param_dict["content"], bos=False, eos=False)
param_dict["content"], bos=False, eos=False
)
else:
content = parser.extract_content_ids(output_tokens)
assert content == []