Upload folder using huggingface_hub

README.md

@@ -0,0 +1,80 @@
+---
+library_name: transformers
+pipeline_tag: automatic-speech-recognition
+inference: true
+---
+
+This model is for debugging. It is randomly initialized with the config from [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) but is of smaller size. 
+
+Codes:
+```python
+import os
+
+import torch
+
+from huggingface_hub import create_repo, upload_folder
+from transformers import (
+    AutoModelForCausalLM,
+    AutoTokenizer,
+    GenerationConfig,
+    AutoConfig,
+    pipeline,
+    set_seed,
+)
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline, AutoConfig
+from datasets import load_dataset
+
+model_id = "openai/whisper-large-v3"
+repo_id = "yujiepan/whisper-v3-tiny-random"
+save_path = f"/tmp/{repo_id}"
+os.system(f'rm -rf {save_path}')
+os.makedirs(save_path, exist_ok=True)
+
+device = "cuda"
+torch_dtype = torch.float16
+model_id = "openai/whisper-large-v3"
+
+config = AutoConfig.from_pretrained(model_id)
+config.num_hidden_layers = 2
+config.d_model = 8
+config.decoder_attention_heads = 2
+config.decoder_ffn_dim = 16
+config.decoder_layers = 2
+config.encoder_ffn_dim = 16
+config.encoder_attention_heads = 2
+config.encoder_layers = 2
+
+model = AutoModelForSpeechSeq2Seq.from_config(config)
+model.to(device).to(torch_dtype)
+model.generation_config = GenerationConfig.from_pretrained(model_id)
+processor = AutoProcessor.from_pretrained(model_id)
+
+set_seed(42)
+num_params = 0
+with torch.no_grad():
+    for name, p in sorted(model.named_parameters()):
+        print(name, p.shape)
+        torch.nn.init.uniform_(p, -0.5, 0.5)
+        num_params += p.numel()
+print("Total number of parameters:", num_params)
+
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model=model,
+    tokenizer=processor.tokenizer,
+    feature_extractor=processor.feature_extractor,
+    torch_dtype=torch_dtype,
+    device=device,
+)
+
+sample = load_dataset(
+    "distil-whisper/librispeech_long", "clean",
+    split="validation",
+)[0]["audio"]
+result = pipe(sample, return_timestamps=True)
+print(result["text"])
+
+create_repo(repo_id, exist_ok=True)
+upload_folder(repo_id=repo_id, folder_path=save_path, repo_type='model')
+```