[Frontend] support matryoshka representation / support embedding API dimensions (#16331)

This commit is contained in:
wang.yuqi
2025-04-11 23:23:10 -07:00
committed by GitHub
parent e92d7085bf
commit fbf722c6e6
11 changed files with 253 additions and 22 deletions
+8 -8
View File
@@ -960,19 +960,19 @@ class VllmRunner:
req_outputs = self.model.classify(prompts)
return [req_output.outputs.probs for req_output in req_outputs]
def encode(
self,
prompts: list[str],
images: Optional[PromptImageInput] = None,
videos: Optional[PromptVideoInput] = None,
audios: Optional[PromptAudioInput] = None,
) -> list[list[float]]:
def encode(self,
prompts: list[str],
images: Optional[PromptImageInput] = None,
videos: Optional[PromptVideoInput] = None,
audios: Optional[PromptAudioInput] = None,
*args,
**kwargs) -> list[list[float]]:
inputs = self.get_inputs(prompts,
images=images,
videos=videos,
audios=audios)
req_outputs = self.model.embed(inputs)
req_outputs = self.model.embed(inputs, *args, **kwargs)
return [req_output.outputs.embedding for req_output in req_outputs]
def score(