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

[1/N] torch.compile user interface design (#10237)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-11 18:01:06 -08:00
committed by GitHub
parent 9cdba9669c
commit eea55cca5b
4 changed files with 55 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
vllm/config.py
View File

@@ -2041,12 +2041,15 @@ class VllmConfig:
simplifies passing around the distinct configurations in the codebase.
"""
model_config: ModelConfig
cache_config: CacheConfig
parallel_config: ParallelConfig
scheduler_config: SchedulerConfig
device_config: DeviceConfig
load_config: LoadConfig
model_config: ModelConfig = field(default=None, init=True) # type: ignore
cache_config: CacheConfig = field(default=None, init=True) # type: ignore
parallel_config: ParallelConfig = field(default=None,
init=True) # type: ignore
scheduler_config: SchedulerConfig = field(default=None,
init=True) # type: ignore
device_config: DeviceConfig = field(default=None,
init=True) # type: ignore
load_config: LoadConfig = field(default=None, init=True) # type: ignore
lora_config: Optional[LoRAConfig] = None
speculative_config: Optional[SpeculativeConfig] = None
decoding_config: Optional[DecodingConfig] = None
@@ -2091,11 +2094,14 @@ class VllmConfig:
def __post_init__(self):
"""Verify configs are valid & consistent with each other.
"""
self.model_config.verify_async_output_proc(self.parallel_config,
self.speculative_config,
self.device_config)
self.model_config.verify_with_parallel_config(self.parallel_config)
self.cache_config.verify_with_parallel_config(self.parallel_config)
if self.model_config is not None:
self.model_config.verify_async_output_proc(self.parallel_config,
self.speculative_config,
self.device_config)
self.model_config.verify_with_parallel_config(self.parallel_config)
if self.cache_config is not None:
self.cache_config.verify_with_parallel_config(self.parallel_config)
if self.lora_config:
self.lora_config.verify_with_model_config(self.model_config)
@@ -2149,4 +2155,4 @@ class VllmConfig:
self.scheduler_config.num_scheduler_steps,
self.cache_config.enable_prefix_caching,
self.model_config.use_async_output_proc,
self.model_config.mm_processor_kwargs)
self.model_config.mm_processor_kwargs)