xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 09:01:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Support FIPS enabled machines with MD5 hashing (#15299)

Browse Source 
Signed-off-by: Matthew Vine <32849887+MattTheCuber@users.noreply.github.com>
This commit is contained in:
Matthew Vine 2025-03-26 20:19:46 -04:00 committed by GitHub
parent e74ff409e0
commit 7a6d45bc8a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 36 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tests/compile/piecewise/test_toy_llama.py
View File

@ -63,7 +63,8 @@ class LlamaConfig:
factors.append((k, v)) factors.append((k, v))
factors.sort() factors.sort()
import hashlib import hashlib
return hashlib.md5(str(factors).encode()).hexdigest() return hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
def __post_init__(self): def __post_init__(self):
assert self.mlp_size >= self.hidden_size assert self.mlp_size >= self.hidden_size

7
vllm/compilation/backends.py
View File

@ -381,8 +381,8 @@ class VllmBackend:
with open(filepath) as f: with open(filepath) as f:
hash_content.append(f.read()) hash_content.append(f.read())
import hashlib import hashlib
code_hash = hashlib.md5( code_hash = hashlib.md5("\n".join(hash_content).encode(),
"\n".join(hash_content).encode()).hexdigest() usedforsecurity=False).hexdigest()
factors.append(code_hash) factors.append(code_hash)
# 3. compiler hash # 3. compiler hash
@ -390,7 +390,8 @@ class VllmBackend:
factors.append(compiler_hash) factors.append(compiler_hash)
# combine all factors to generate the cache dir # combine all factors to generate the cache dir
hash_key = hashlib.md5(str(factors).encode()).hexdigest()[:10] hash_key = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()[:10]
cache_dir = os.path.join( cache_dir = os.path.join(
envs.VLLM_CACHE_ROOT, envs.VLLM_CACHE_ROOT,

3
vllm/compilation/compiler_interface.py
View File

@ -139,7 +139,8 @@ class InductorAdaptor(CompilerInterface):
from torch._inductor.codecache import torch_key from torch._inductor.codecache import torch_key
torch_factors = torch_key() torch_factors = torch_key()
factors.append(torch_factors) factors.append(torch_factors)
hash_str = hashlib.md5(str(factors).encode()).hexdigest()[:10] hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()[:10]
return hash_str return hash_str
def initialize_cache(self, cache_dir: str, disable_cache: bool = False): def initialize_cache(self, cache_dir: str, disable_cache: bool = False):

42
vllm/config.py
View File

@ -1111,7 +1111,8 @@ class CacheConfig:
factors: list[Any] = [] factors: list[Any] = []
factors.append(self.cache_dtype) factors.append(self.cache_dtype)
# `cpu_offload_gb` does not use `torch.compile` yet. # `cpu_offload_gb` does not use `torch.compile` yet.
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __init__( def __init__(
@ -1243,7 +1244,8 @@ class TokenizerPoolConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -1354,7 +1356,8 @@ class LoadConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -1674,7 +1677,8 @@ class SchedulerConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self) -> None: def __post_init__(self) -> None:
@ -1810,7 +1814,8 @@ class DeviceConfig:
# the device/platform information will be summarized # the device/platform information will be summarized
# by torch/vllm automatically. # by torch/vllm automatically.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __init__(self, device: str = "auto") -> None: def __init__(self, device: str = "auto") -> None:
@ -1983,7 +1988,8 @@ class SpeculativeConfig:
# no factors to consider. # no factors to consider.
# spec decode does not use `torch.compile` yet. # spec decode does not use `torch.compile` yet.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
@classmethod @classmethod
@ -2358,7 +2364,8 @@ class LoRAConfig:
factors.append(self.lora_extra_vocab_size) factors.append(self.lora_extra_vocab_size)
factors.append(self.long_lora_scaling_factors) factors.append(self.long_lora_scaling_factors)
factors.append(self.bias_enabled) factors.append(self.bias_enabled)
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -2424,7 +2431,8 @@ class PromptAdapterConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -2469,7 +2477,8 @@ class MultiModalConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def get_limit_per_prompt(self, modality: str) -> int: def get_limit_per_prompt(self, modality: str) -> int:
@ -2535,7 +2544,8 @@ class PoolerConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
@staticmethod @staticmethod
@ -2816,7 +2826,8 @@ class DecodingConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -2866,7 +2877,8 @@ class ObservabilityConfig:
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
def __post_init__(self): def __post_init__(self):
@ -2928,7 +2940,8 @@ class KVTransferConfig(BaseModel):
# no factors to consider. # no factors to consider.
# this config will not affect the computation graph. # this config will not affect the computation graph.
factors: list[Any] = [] factors: list[Any] = []
hash_str = hashlib.md5(str(factors).encode()).hexdigest() hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()
return hash_str return hash_str
@classmethod @classmethod
@ -3425,7 +3438,8 @@ class VllmConfig:
vllm_factors.append("None") vllm_factors.append("None")
factors.append(vllm_factors) factors.append(vllm_factors)
hash_str = hashlib.md5(str(factors).encode()).hexdigest()[:10] hash_str = hashlib.md5(str(factors).encode(),
usedforsecurity=False).hexdigest()[:10]
return hash_str return hash_str
def pad_for_cudagraph(self, batch_size: int) -> int: def pad_for_cudagraph(self, batch_size: int) -> int: