initial commit

app.py

@@ -28,7 +28,7 @@ os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
 api_key = os.getenv("my_secret")
 
-snapshot_download(repo_id="nvidia/audio-flamingo-2-1.5B", local_dir="./", token=api_key)
+snapshot_download(repo_id="nvidia/audio-flamingo-2-0.5B", local_dir="./", token=api_key)
 
 config = yaml.load(open("configs/inference.yaml"), Loader=yaml.FullLoader)
 

configs/inference.yaml

@@ -78,8 +78,8 @@ mert_config:
 model_config:
   cache_dir: .cache
 
-  lang_encoder_path: Qwen/Qwen2.5-1.5B
-  tokenizer_path: Qwen/Qwen2.5-1.5B
+  lang_encoder_path: Qwen/Qwen2.5-0.5B
+  tokenizer_path: Qwen/Qwen2.5-0.5B
   cross_attn_every_n_layers: 1
   audio_transformer_kwargs: {
     n_head: 8,