Updating lm-format-enforcer version and adding links to decoding libraries in docs (#4222)

requirements-common.txt

@@ -12,7 +12,7 @@ uvicorn[standard]
 pydantic >= 2.0  # Required for OpenAI server.
 prometheus_client >= 0.18.0
 tiktoken == 0.6.0  # Required for DBRX tokenizer
-lm-format-enforcer == 0.9.3
+lm-format-enforcer == 0.9.8
 outlines == 0.0.34 # Requires torch >= 2.1.0
 typing_extensions
 filelock >= 3.10.4 # filelock starts to support `mode` argument from 3.10.4

vllm/engine/arg_utils.py

@@ -197,7 +197,11 @@ class EngineArgs:
             default='outlines',
             choices=['outlines', 'lm-format-enforcer'],
             help='Which engine will be used for guided decoding'
-            ' (JSON schema / regex etc).')
+            ' (JSON schema / regex etc) by default. Currently support '
+            'https://github.com/outlines-dev/outlines and '
+            'https://github.com/noamgat/lm-format-enforcer.'
+            ' Can be overridden per request via guided_decoding_backend'
+            ' parameter.')
         # Parallel arguments
         parser.add_argument('--worker-use-ray',
                             action='store_true',

vllm/model_executor/layers/quantization/fp8.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, Tuple
 
 import torch
 from torch.nn import Module
@@ -114,7 +114,7 @@ class Fp8LinearMethod(LinearMethodBase):
         return output
 
 
-def per_tensor_quantize(tensor: torch.Tensor) -> tuple[torch.Tensor, float]:
+def per_tensor_quantize(tensor: torch.Tensor) -> Tuple[torch.Tensor, float]:
     """Quantize a tensor using per-tensor static scaling factor.
 
     Args: