using this model over 0.6b??
#2
by ineednot1 - opened
tokenizer = Qwen2TokenizerFast.from_pretrained("ACE-Step/acestep-5Hz-lm-4B")
text_encoder = Qwen3Model.from_pretrained("ACE-Step/acestep-5Hz-lm-4B")
pipe = AceStepPipeline.from_pretrained(
"ACE-Step/acestep-v15-xl-turbo-diffusers",
torch_dtype=torch.bfloat16,
)
pipe.text_encoder=text_encoder
pipe.tokenizer=tokenizer
RuntimeError: mat1 and mat2 shapes cannot be multiplied (64x2560 and 1024x2048)
any other way?