GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
LakoreAI
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
LakoreAI
Model Tree
openai/whisper-small
Input Modalities
Output Modalities
Supported Functionality
| Metric | Value |
|---|---|
| Training Loss | 0.9382 |
| Epochs | 5 |
| Global Steps | 470 |
| Samples/sec | 7.37 |
| Total FLOPs | 4.60e+18 |
openai/whisper-small (244M params)q_proj, v_proj, k_proj, out_proj, fc1, fc2vi)vi)Data augmentation applied:
python
from peft import PeftModelfrom transformers import WhisperForConditionalGeneration, WhisperProcessorimport torch base = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")model = PeftModel.from_pretrained(base, "LakoreAI/whisper-small-vi-lora")processor = WhisperProcessor.from_pretrained("LakoreAI/whisper-small-vi-lora") # Optional: merge LoRA for faster inferencemodel = model.merge_and_unload()model.eval() # Inferencedef transcribe(audio_array, sampling_rate=16000): inputs = processor(audio_array, sampling_rate=sampling_rate, return_tensors="pt") with torch.no_grad(): ids = model.generate( inputs.input_features, language="vietnamese", task="transcribe", max_new_tokens=225, ) return processor.tokenizer.decode(ids[0], skip_special_tokens=True)from peft import PeftModelfrom transformers import WhisperForConditionalGeneration, WhisperProcessorimport torch base = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")model = PeftModel.from_pretrained(base, "LakoreAI/whisper-small-vi-lora")processor = WhisperProcessor.from_pretrained("LakoreAI/whisper-small-vi-lora") # Optional: merge LoRA for faster inferencemodel = model.merge_and_unload()model.eval() # Inferencedef transcribe(audio_array, sampling_rate=16000): inputs = processor(audio_array, sampling_rate=sampling_rate, return_tensors="pt") with torch.no_grad(): ids = model.generate( inputs.input_features, language="vietnamese", task="transcribe", max_new_tokens=225, ) return processor.tokenizer.decode(ids[0], skip_special_tokens=True)