GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
maryamas222
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
maryamas222
Model Tree
openai/whisper-large-v3
Input Modalities
Output Modalities
Supported Functionality
| Property | Value |
|---|---|
| Base model | openai/whisper-large-v3 |
| Fine-tuning method | LoRA (PEFT) |
| Language | Egyptian Arabic |
| Task | Automatic Speech Recognition |
| Training dataset | MAdel121/arabic-egy-cleaned |
| Training samples | ~82,900 |
| LoRA rank | 32 |
| LoRA alpha | 64 |
| LoRA target modules | q_proj, v_proj, k_proj, out_proj |
| Metric | Score |
|---|---|
| WER (arabic-egy-cleaned test split) | 36.92% |
Evaluation was performed with Arabic text normalization applied (alef variants normalized).
python
import torchfrom transformers import WhisperProcessor, WhisperForConditionalGenerationfrom peft import PeftModel BASE_MODEL_ID = "openai/whisper-large-v3"LORA_MODEL_ID = "maryamas222/whisper-large-v3-egyptian-lora-v4" processor = WhisperProcessor.from_pretrained(BASE_MODEL_ID, language="Arabic", task="transcribe") base_model = WhisperForConditionalGeneration.from_pretrained( BASE_MODEL_ID, torch_dtype=torch.float16, device_map="auto")model = PeftModel.from_pretrained(base_model, LORA_MODEL_ID)model = model.merge_and_unload()model.eval() # Transcribe audioinputs = processor(audio_array, sampling_rate=16000, return_tensors="pt").input_featuresinputs = inputs.to("cuda").to(torch.float16) with torch.no_grad(): predicted_ids = model.generate( inputs, language="arabic", task="transcribe", max_new_tokens=225, ) transcription = processor.tokenizer.decode(predicted_ids[0], skip_special_tokens=True)print(transcription)Training samples were filtered to:
Alef variants (إأآ) normalized to (ا) for consistency between training labels and evaluation.
This model is intended for academic research and experimentation with Egyptian Arabic ASR. It is not recommended for production use without further evaluation on domain-specific data.
bibtex
@misc{radford2022whisper, title={Robust Speech Recognition via Large-Scale Weak Supervision}, author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya}, year={2022}}import torchfrom transformers import WhisperProcessor, WhisperForConditionalGenerationfrom peft import PeftModel BASE_MODEL_ID = "openai/whisper-large-v3"LORA_MODEL_ID = "maryamas222/whisper-large-v3-egyptian-lora-v4" processor = WhisperProcessor.from_pretrained(BASE_MODEL_ID, language="Arabic", task="transcribe") base_model = WhisperForConditionalGeneration.from_pretrained( BASE_MODEL_ID, torch_dtype=torch.float16, device_map="auto")model = PeftModel.from_pretrained(base_model, LORA_MODEL_ID)model = model.merge_and_unload()model.eval() # Transcribe audioinputs = processor(audio_array, sampling_rate=16000, return_tensors="pt").input_featuresinputs = inputs.to("cuda").to(torch.float16) with torch.no_grad(): predicted_ids = model.generate( inputs, language="arabic", task="transcribe", max_new_tokens=225, ) transcription = processor.tokenizer.decode(predicted_ids[0], skip_special_tokens=True)print(transcription)@misc{radford2022whisper, title={Robust Speech Recognition via Large-Scale Weak Supervision}, author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya}, year={2022}}