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

[misc][core] lazy import outlines (#7831)

Browse Source
This commit is contained in:
youkaichao
2024-08-24 00:51:38 -07:00
committed by GitHub
parent d81abefd2e
commit 7d9ffa2ae1
4 changed files with 64 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/guided_decoding/__init__.py
View File

@@ -5,9 +5,6 @@ from vllm.entrypoints.openai.protocol import (
CompletionRequest)
from vllm.model_executor.guided_decoding.guided_fields import (
GuidedDecodingRequest)
from vllm.model_executor.guided_decoding.outlines_decoding import (
get_local_outlines_guided_decoding_logits_processor,
get_outlines_guided_decoding_logits_processor)
from vllm.sampling_params import LogitsProcessor
@@ -18,6 +15,9 @@ async def get_guided_decoding_logits_processor(
request = _adapt_request_for_tool_use(request)
if guided_decoding_backend == 'outlines':
# NOTE: lazy import outlines to avoid https://github.com/vllm-project/vllm/issues/4193
from vllm.model_executor.guided_decoding.outlines_decoding import ( # noqa
get_outlines_guided_decoding_logits_processor)
return await get_outlines_guided_decoding_logits_processor(
request, tokenizer)
if guided_decoding_backend == 'lm-format-enforcer':
@@ -37,6 +37,9 @@ def get_local_guided_decoding_logits_processor(
# request = _adapt_request_for_tool_use(request)
if guided_decoding_backend == 'outlines':
# NOTE: lazy import outlines to avoid https://github.com/vllm-project/vllm/issues/4193
from vllm.model_executor.guided_decoding.outlines_decoding import ( # noqa
get_local_outlines_guided_decoding_logits_processor)
return get_local_outlines_guided_decoding_logits_processor(
guided_options, tokenizer)
if guided_decoding_backend == 'lm-format-enforcer':

11
vllm/model_executor/guided_decoding/lm_format_enforcer_decoding.py
View File

@@ -14,9 +14,6 @@ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
CompletionRequest)
from vllm.model_executor.guided_decoding.guided_fields import (
GuidedDecodingRequest)
from vllm.model_executor.guided_decoding.outlines_decoding import (
get_local_outlines_guided_decoding_logits_processor,
get_outlines_guided_decoding_logits_processor)
from vllm.sampling_params import LogitsProcessor
@@ -43,6 +40,10 @@ async def get_lm_format_enforcer_guided_decoding_logits_processor(
character_level_parser = RegexParser(request.guided_regex)
elif request.guided_grammar:
# CFG grammar not supported by LMFE, revert to outlines
# NOTE: lazy import outlines to avoid https://github.com/vllm-project/vllm/issues/4193
from vllm.model_executor.guided_decoding.outlines_decoding import (
get_outlines_guided_decoding_logits_processor)
return await get_outlines_guided_decoding_logits_processor(
request, tokenizer)
elif (request.response_format is not None
@@ -87,6 +88,10 @@ def get_local_lm_format_enforcer_guided_decoding_logits_processor(
character_level_parser = RegexParser(guided_options.guided_regex)
elif guided_options.guided_grammar:
# CFG grammar not supported by LMFE, revert to outlines
# NOTE: lazy import outlines to avoid https://github.com/vllm-project/vllm/issues/4193
from vllm.model_executor.guided_decoding.outlines_decoding import (
get_local_outlines_guided_decoding_logits_processor)
return get_local_outlines_guided_decoding_logits_processor(
guided_options, tokenizer)
elif guided_options.guided_json_object: