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

[CI] improve embed testing (#18747)

Browse Source
This commit is contained in:
wang.yuqi
2025-05-28 15:16:35 +08:00
committed by GitHub
parent 0c492b7824
commit de65fc8e1e
13 changed files with 248 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm/config.py
View File

@@ -572,13 +572,7 @@ class ModelConfig:
sliding_window = None
self.original_max_model_len = self.max_model_len
self.max_model_len = _get_and_verify_max_len(
hf_config=self.hf_text_config,
max_model_len=self.max_model_len,
disable_sliding_window=self.disable_sliding_window,
sliding_window_len=self.get_hf_config_sliding_window(),
spec_target_max_model_len=self.spec_target_max_model_len,
encoder_config=self.encoder_config)
self.max_model_len = self.get_and_verify_max_len(self.max_model_len)
self.served_model_name = get_served_model_name(self.model,
self.served_model_name)
self.multimodal_config = self._init_multimodal_config()
@@ -1382,6 +1376,16 @@ class ModelConfig:
def matryoshka_dimensions(self):
return getattr(self.hf_config, "matryoshka_dimensions", None)
def get_and_verify_max_len(self, max_model_len: int):
max_model_len = _get_and_verify_max_len(
hf_config=self.hf_text_config,
max_model_len=max_model_len,
disable_sliding_window=self.disable_sliding_window,
sliding_window_len=self.get_hf_config_sliding_window(),
spec_target_max_model_len=self.spec_target_max_model_len,
encoder_config=self.encoder_config)
return max_model_len
BlockSize = Literal[1, 8, 16, 32, 64, 128]
CacheDType = Literal["auto", "fp8", "fp8_e4m3", "fp8_e5m2"]
@@ -4469,13 +4473,7 @@ class VllmConfig:
def recalculate_max_model_len(self, max_model_len: int):
model_config = self.model_config
max_model_len = _get_and_verify_max_len(
hf_config=model_config.hf_text_config,
max_model_len=max_model_len,
disable_sliding_window=model_config.disable_sliding_window,
sliding_window_len=model_config.get_hf_config_sliding_window(),
spec_target_max_model_len=model_config.spec_target_max_model_len,
encoder_config=model_config.encoder_config)
max_model_len = model_config.get_and_verify_max_len(max_model_len)
self.model_config.max_model_len = max_model_len
self.scheduler_config.max_model_len = max_model_len
self.compute_hash()