Removes duplicate outlines processors

tests/entrypoints/openai/test_guided_processors.py

@@ -2,18 +2,17 @@
 # tests/test_guided_decoding directory.
 import pytest
 import torch
+from outlines.processors import JSONLogitsProcessor, RegexLogitsProcessor
 from transformers import AutoTokenizer
 
 from vllm.entrypoints.openai.protocol import CompletionRequest
 from vllm.model_executor.guided_decoding import (
     get_guided_decoding_logits_processor)
-from vllm.model_executor.guided_decoding.outlines_logits_processors import (
-    JSONLogitsProcessor, RegexLogitsProcessor)
 
 
 def test_guided_logits_processors(sample_regex, sample_json_schema):
     """Basic unit test for RegexLogitsProcessor and JSONLogitsProcessor."""
-    tokenizer = AutoTokenizer.from_pretrained('HuggingFaceH4/zephyr-7b-beta')
+    tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta", )
     regex_LP = RegexLogitsProcessor(sample_regex, tokenizer)
     json_LP = JSONLogitsProcessor(sample_json_schema,
                                   tokenizer,
@@ -41,10 +40,10 @@ def test_guided_logits_processors(sample_regex, sample_json_schema):
 @pytest.mark.parametrize("backend", ["outlines", "lm-format-enforcer"])
 async def test_guided_logits_processor_black_box(backend: str, sample_regex,
                                                  sample_json_schema):
-    tokenizer = AutoTokenizer.from_pretrained('HuggingFaceH4/zephyr-7b-beta')
+    tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
     token_ids = tokenizer.encode(
         f"Give an example IPv4 address with this regex: {sample_regex}")
-    regex_request = CompletionRequest(model='test',
+    regex_request = CompletionRequest(model="test",
                                       prompt=token_ids,
                                       guided_regex=sample_regex)
     regex_lp = await get_guided_decoding_logits_processor(
@@ -59,7 +58,7 @@ async def test_guided_logits_processor_black_box(backend: str, sample_regex,
     token_ids = tokenizer.encode(
         f"Give an employee profile that fits this schema: {sample_json_schema}"
     )
-    json_request = CompletionRequest(model='test',
+    json_request = CompletionRequest(model="test",
                                      prompt=token_ids,
                                      guided_json=sample_json_schema)
     json_lp = await get_guided_decoding_logits_processor(

vllm/model_executor/guided_decoding/outlines_decoding.py

@@ -5,13 +5,13 @@
 from re import escape as regex_escape
 from typing import Tuple, Union
 
+from outlines.processors import (CFGLogitsProcessor, JSONLogitsProcessor,
+                                 RegexLogitsProcessor)
 from pydantic import BaseModel
 from transformers import PreTrainedTokenizerBase
 
 from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
                                               CompletionRequest)
-from vllm.model_executor.guided_decoding.outlines_logits_processors import (
-    CFGLogitsProcessor, JSONLogitsProcessor, RegexLogitsProcessor)
 
 
 class GuidedDecodingMode(Enum):
@@ -52,8 +52,8 @@ class GuidedDecodingMode(Enum):
 
 
 async def get_outlines_guided_decoding_logits_processor(
-    request: Union[CompletionRequest,
-                   ChatCompletionRequest], tokenizer: PreTrainedTokenizerBase
+    request: Union[CompletionRequest, ChatCompletionRequest],
+    tokenizer: PreTrainedTokenizerBase,
 ) -> Union[JSONLogitsProcessor, RegexLogitsProcessor, CFGLogitsProcessor,
            None]:
     """
@@ -72,9 +72,14 @@ async def get_outlines_guided_decoding_logits_processor(
             max_workers=2)
     loop = asyncio.get_running_loop()
 
-    return await loop.run_in_executor(global_thread_pool,
-                                      _get_logits_processor, guide, tokenizer,
-                                      mode, request.guided_whitespace_pattern)
+    return await loop.run_in_executor(
+        global_thread_pool,
+        _get_logits_processor,
+        guide,
+        tokenizer,
+        mode,
+        request.guided_whitespace_pattern,
+    )
 
 
 def _get_guide_and_mode(
@@ -110,8 +115,10 @@ def _get_guide_and_mode(
 
 
 def _get_logits_processor(
-    guide: str, tokenizer: PreTrainedTokenizerBase, mode: GuidedDecodingMode,
-    whitespace_pattern: Union[str, None]
+    guide: str,
+    tokenizer: PreTrainedTokenizerBase,
+    mode: GuidedDecodingMode,
+    whitespace_pattern: Union[str, None],
 ) -> Union[JSONLogitsProcessor, RegexLogitsProcessor, CFGLogitsProcessor]:
     if mode == GuidedDecodingMode.JSON:
         return JSONLogitsProcessor(guide, tokenizer, whitespace_pattern)