GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
amn-raw
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
amn-raw
Model Tree
openai/whisper-large-v3
Input Modalities
Output Modalities
Supported Functionality
| Property | Value |
|---|---|
| Base model | openai/whisper-large-v3 |
| Parameters | ~1.55B |
| Architecture | Encoder-decoder Transformer |
| Encoder layers | 32 |
| Decoder layers | 32 |
| Hidden size | 1280 |
| Attention heads | 20 |
| Mel bins | 128 |
| Max audio length | 30 seconds (chunked for longer inputs) |
python
from transformers import WhisperForConditionalGeneration, WhisperProcessorimport torch model_id = "amn-raw/naijaVox2-quantized-int8-gguf" processor = WhisperProcessor.from_pretrained(model_id)model = WhisperForConditionalGeneration.from_pretrained( model_id, torch_dtype=torch.float16, device_map="auto") # Transcribefrom datasets import load_dataset, Audiods = load_dataset("your_dataset", split="test")ds = ds.cast_column("audio", Audio(sampling_rate=16000)) sample = ds[0]["audio"]input_features = processor( sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features.to(model.device, dtype=torch.float16) predicted_ids = model.generate(input_features)transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]print(transcription)bash
pip install vllmvllm serve amn-raw/naijaVox2-quantized-int8-gguf \ --max-model-len 448 \ --max-num-seqs 100 \ --task transcriptionThen call the transcription endpoint:
python
from openai import OpenAI client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")response = client.audio.transcriptions.create( model="amn-raw/naijaVox2-quantized-int8-gguf", file=open("audio.wav", "rb"),)print(response.text)Pre-quantized GGUF files are available in this repo under the gguf/ directory.
bash
# Using transcribe.cppbuild/bin/transcribe-cli \ -m naijavox-2.0-Q8_0.gguf \ audio.wav| File | Quant | Size | Notes |
|---|---|---|---|
gguf/naijavox-F16.gguf | F16 | ~2.95 GB | Full precision reference |
gguf/naijavox-2-Q8_0.gguf | Q8_0 | ~1.6 GB | Recommended for most use |
Apache 2.0 (same as base whisper-large-v3)
from transformers import WhisperForConditionalGeneration, WhisperProcessorimport torch model_id = "amn-raw/naijaVox2-quantized-int8-gguf" processor = WhisperProcessor.from_pretrained(model_id)model = WhisperForConditionalGeneration.from_pretrained( model_id, torch_dtype=torch.float16, device_map="auto") # Transcribefrom datasets import load_dataset, Audiods = load_dataset("your_dataset", split="test")ds = ds.cast_column("audio", Audio(sampling_rate=16000)) sample = ds[0]["audio"]input_features = processor( sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features.to(model.device, dtype=torch.float16) predicted_ids = model.generate(input_features)transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]print(transcription)pip install vllmvllm serve amn-raw/naijaVox2-quantized-int8-gguf \ --max-model-len 448 \ --max-num-seqs 100 \ --task transcriptionfrom openai import OpenAI client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")response = client.audio.transcriptions.create( model="amn-raw/naijaVox2-quantized-int8-gguf", file=open("audio.wav", "rb"),)print(response.text)# Using transcribe.cppbuild/bin/transcribe-cli \ -m naijavox-2.0-Q8_0.gguf \ audio.wav