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 16, 22:00 UTC
[Misc] refactor(config): clean up unused code (#6320)
aniaan committed
Jul 11, 2024 at 17:39 UTC
3963a5335bb4106f2ecd1139527e3568d2151933
2 files changed
+2
-7
vllm/config.py
+2
-4
@@ -138,12 +138,10 @@ class ModelConfig:
138
self.quantization = quantization
139
self.quantization_param_path = quantization_param_path
140
self.enforce_eager = enforce_eager
141
- self.max_context_len_to_capture = max_context_len_to_capture
142
- if self.max_context_len_to_capture is not None:
141
+ if max_context_len_to_capture is not None:
142
raise ValueError("`max_context_len_to_capture` is deprecated. "
143
"Use `max_seq_len_to_capture` instead.")
145
- self.max_seq_len_to_capture = (max_seq_len_to_capture
146
- or max_context_len_to_capture)
144
+ self.max_seq_len_to_capture = max_seq_len_to_capture
145
self.max_logprobs = max_logprobs
146
self.disable_sliding_window = disable_sliding_window
147
self.skip_tokenizer_init = skip_tokenizer_init
vllm/worker/xpu_model_runner.py
-3
@@ -109,9 +109,6 @@ class XPUModelRunner(ModelRunnerBase[ModelInputForXPU]):
109
110
self.kv_cache_dtype = kv_cache_dtype
111
self.block_size = cache_config.block_size
112
- self.max_context_len_to_capture = (
113
- self.model_config.max_context_len_to_capture
114
- if self.model_config is not None else 0)
112
113
self.attn_backend = get_attn_backend(
114
self.model_config.get_num_attention_heads(self.parallel_config),