A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 08:30 UTC
Remove checks for `None` for fields which should never be `None` (#17985)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
May 28, 2025 at 22:32 UTC
6dbe5b5c934268387fb03485a5625b1793049d8c
1 file changed
+10
-14
vllm/config.py
+10
-14
@@ -4282,25 +4282,22 @@ class VllmConfig:
4282
self.model_config.verify_dual_chunk_attention_config(
4283
self.load_config)
4284
4285
- if self.cache_config is not None:
4286
- self.cache_config.verify_with_parallel_config(self.parallel_config)
4285
+ self.cache_config.verify_with_parallel_config(self.parallel_config)
4286
4288
- if self.lora_config:
4287
+ if self.lora_config is not None:
4288
self.lora_config.verify_with_cache_config(self.cache_config)
4289
self.lora_config.verify_with_model_config(self.model_config)
4290
self.lora_config.verify_lora_support()
4292
- if self.prompt_adapter_config:
4291
+ if self.prompt_adapter_config is not None:
4292
self.prompt_adapter_config.verify_with_model_config(
4293
self.model_config)
4294
4296
- if self.quant_config is None and \
4297
- self.model_config is not None and self.load_config is not None:
4295
+ if self.quant_config is None and self.model_config is not None:
4296
self.quant_config = VllmConfig._get_quantization_config(
4297
self.model_config, self.load_config)
4298
4299
from vllm.platforms import current_platform
4302
- if self.scheduler_config is not None and \
4303
- self.model_config is not None and \
4300
+ if self.model_config is not None and \
4301
self.scheduler_config.chunked_prefill_enabled and \
4302
self.model_config.dtype == torch.float32 and \
4303
current_platform.get_device_capability() == (7, 5):
@@ -4335,8 +4332,7 @@ class VllmConfig:
4332
4333
self._set_cudagraph_sizes()
4334
4338
- if self.cache_config is not None and \
4339
- self.cache_config.cpu_offload_gb > 0 and \
4335
+ if self.cache_config.cpu_offload_gb > 0 and \
4336
self.compilation_config.level != CompilationLevel.NO_COMPILATION \
4337
and not envs.VLLM_USE_V1:
4338
logger.warning(
@@ -4358,16 +4354,16 @@ class VllmConfig:
4354
"full_cuda_graph is not supported with "
4355
"cascade attention. Disabling cascade attention.")
4356
self.model_config.disable_cascade_attn = True
4361
- if self.cache_config is not None:
4362
- self.cache_config.enable_prefix_caching = False
4357
+ self.cache_config.enable_prefix_caching = False
4358
4364
- if (self.kv_events_config
4359
+ if (self.kv_events_config is not None
4360
and self.kv_events_config.enable_kv_cache_events
4361
and not self.cache_config.enable_prefix_caching):
4362
logger.warning(
4363
"KV cache events are on, but prefix caching is not enabled."
4364
"Use --enable-prefix-caching to enable.")
4370
- if (self.kv_events_config and self.kv_events_config.publisher != "null"
4365
+ if (self.kv_events_config is not None
4366
+ and self.kv_events_config.publisher != "null"
4367
and not self.kv_events_config.enable_kv_cache_events):
4368
logger.warning("KV cache events are disabled,"
4369
"but the scheduler is configured to publish them."