Update to new version of base image

vllm/model_executor/guided_decoding/outlines_logits_processors.py

@@ -23,12 +23,21 @@ from typing import Callable, DefaultDict, Dict, List, Union
 import torch
 from lark import Lark
 from outlines import grammars
-from outlines.caching import cache
+from outlines.caching import cache, disable_cache
 from outlines.fsm.guide import CFGGuide, Generate, Guide, RegexGuide, Write
 from outlines.fsm.json_schema import build_regex_from_schema
 from pydantic import BaseModel
 from transformers import PreTrainedTokenizerBase
+import vllm.envs as envs
+from vllm.logger import init_logger
+logger = init_logger(__name__)
 
+if envs.VLLM_V0_USE_OUTLINES_CACHE:
+    logger.warning("Enabling outlines cache. This is an unbounded on-disk "
+                   "cache. It may consume a lot of disk space and should "
+                   "not be used with untrusted clients.")
+else:
+    disable_cache()
 
 class BaseLogitsProcessor:
 

vllm/model_executor/layers/linear.py

@@ -132,7 +132,7 @@ class UnquantizedLinearMethod(LinearMethodBase):
               layer: torch.nn.Module,
               x: torch.Tensor,
               bias: Optional[torch.Tensor] = None) -> torch.Tensor:
-        if (x.shape[0] == 16384 or x.shape[0] == 15360):
+        if (x.shape[0] == 8192 or x.shape[0] == 16384 or x.shape[0] == 15360):
             if bias is None:
                 return x @ layer.weight.T
             else:

vllm/model_executor/layers/quantization/compressed_tensors/utils.py

@@ -1,4 +1,4 @@
-import re
+import regex as re
 from enum import Enum
 from typing import Any, Dict, Iterable, Optional, Union
 

vllm/model_executor/model_loader/tensorizer.py

@@ -2,7 +2,7 @@ import argparse
 import dataclasses
 import io
 import os
-import re
+import regex as re
 import time
 from dataclasses import dataclass
 from functools import partial

vllm/model_executor/model_loader/weight_utils.py

@@ -85,7 +85,7 @@ def convert_bin_to_safetensor_file(
     pt_filename: str,
     sf_filename: str,
 ) -> None:
-    loaded = torch.load(pt_filename, map_location="cpu")
+    loaded = torch.load(pt_filename, map_location="cpu", weights_only=True)
     if "state_dict" in loaded:
         loaded = loaded["state_dict"]
     shared = _shared_pointers(loaded)
@@ -373,7 +373,7 @@ def np_cache_weights_iterator(
                     disable=not enable_tqdm,
                     bar_format=_BAR_FORMAT,
             ):
-                state = torch.load(bin_file, map_location="cpu")
+                state = torch.load(bin_file, map_location="cpu", weights_only=True)
                 for name, param in state.items():
                     param_path = os.path.join(np_folder, name)
                     with open(param_path, "wb") as f:
@@ -422,7 +422,7 @@ def pt_weights_iterator(
             disable=not enable_tqdm,
             bar_format=_BAR_FORMAT,
     ):
-        state = torch.load(bin_file, map_location="cpu")
+        state = torch.load(bin_file, map_location="cpu", weights_only=True)
         for name, param in state.items():
             yield name, param
         del state