Spaces:

victor
/

MOSS-TTS-Nano

Running on Zero

victor HF Staff commited on 15 days ago

Commit

749bf48

1 Parent(s): 8ee99c8

Fix: set attn_implementation=sdpa (flash_attn not installed)

Files changed (1) hide show

app.py CHANGED Viewed

@@ -40,6 +40,7 @@ tts_model = AutoModelForCausalLM.from_pretrained(
     low_cpu_mem_usage=True,
 )
 tts_model.eval()
 logger.info("Loading audio tokenizer: %s", AUDIO_TOKENIZER_ID)
 audio_tokenizer = AutoModel.from_pretrained(

     low_cpu_mem_usage=True,
 )
 tts_model.eval()
+tts_model._set_attention_implementation("sdpa")  # flash_attn not installed; sdpa is always available
 logger.info("Loading audio tokenizer: %s", AUDIO_TOKENIZER_ID)
 audio_tokenizer = AutoModel.from_pretrained(