No official torrent yet — requesting it tells the flock what to host next. You cannot seed a catalog-only page.
Cataloged from IFM/Jais-ASR-GGUF
All models by IFM →[!NOTE] This repository contains GGUF versions of IFM/Jais-ASR for use with
llama.cpp. Each configuration uses two files: a text decoder (Jais-ASR-*.gguf) and an audio encoder (mmproj-Jais-ASR-*.gguf); choose one in the table below. The Quickstart further down applies to the original model.Compatibility: requires the MBZUAI-IFM fork of llama.cpp with
llama.cpp-jais-asr.patchfrom this repository.
WER / CER (%) are averaged over the six test sets, evaluated with llama.cpp (greedy decoding, code_switch=False, at most 256 generated tokens); lower is better. Size is the download size of both files (text decoder and audio encoder). GPU memory is the peak on one GPU.
The 8-bit configuration matches the original model's accuracy on the leaderboard (22.83% vs. 22.85% WER) while using about 39% less GPU memory.
Jais-ASR is IFM's state-of-the-art speech recognition model for Arabic as it is spoken: Modern Standard Arabic and dialects from 19 Arab countries, with natural code-switching into English, French, Spanish, and Italian. Based on Jais-3-4B, it adds optional diacritization, built-in language and dialect identification, and an explicit signal when there is no speech to transcribe. The weights are released under Apache 2.0 with a vLLM plugin, and quantized versions are available in the Jais Speech collection.
[Image blocked: Average word and character error rates on the Open Universal Arabic ASR Leaderboard. Jais-ASR has the lowest of both, 22.85% WER and 8.37% CER, ahead of Audar-ASR-V1-Turbo, Speechmatics Melia 1, ElevenLabs Scribe v2, cohere-transcribe-arabic-07-2026, omniASR-LLM-7B, Qwen3-Omni-30B-A3B, Qwen3-ASR-1.7B and Voxtral-Small-24B.]
/v1/audio/transcriptions endpoint. The bf16 weights are 11.4 GB.The Open Universal Arabic ASR Leaderboard covers six public test sets spanning read speech, broadcast media, and conversational dialects, in clean and noisy conditions. Evaluated with the leaderboard's own protocol, Jais-ASR leads on both average metrics.
Error rates in %; lower is better. Average is the macro-average over the six test sets. Jais-ASR was measured with the leaderboard's evaluation script. Scores for the other open-weight models are taken from the leaderboard; Speechmatics Melia 1 and ElevenLabs Scribe v2 were measured by IFM.
Jais-ASR brings Arabic speech transcription, code-switching, diacritization, and language identification into one model, with controls to tailor the output to your application.
code_switch=True to keep English, French, Spanish, or Italian spans in their original script alongside Arabic, useful for interviews and everyday conversations.diacritics=True to add vowel marks to Arabic transcripts, making them easier to read and supporting language learning. Diacritics and code-switching can be enabled together.[IFM-NoTranscript] with the flag refused=True instead of hallucinated content, giving applications a clear signal to handle the case separately.Automatic language and dialect detection is enabled by default. If the language of the audio is known in advance, set it with one of the codes below: ar allows Modern Standard Arabic and all 19 supported dialects, msa selects Modern Standard Arabic, and a dialect code such as EG selects that dialect.
The 19 Arabic dialect codes are:
Algerian (DZ) · Bahraini (BH) · Egyptian (EG) · Emirati (AE) · Iraqi (IQ) · Jordanian (JO) · Kuwaiti (KW) · Lebanese (LB) · Libyan (LY) · Mauritanian (MR) · Moroccan (MA) · Omani (OM) · Palestinian (PS) · Qatari (QA) · Saudi (SA) · Sudanese (SD) · Syrian (SY) · Tunisian (TN) · Yemeni (YE)
Install PyTorch for your hardware, then install the inference dependencies:
The examples below use a GPU with bfloat16 support. The bf16 weights are 11.4 GB; plan on roughly 12 GB of GPU memory.
The pipeline accepts an audio file path or a waveform dictionary such as {"array": waveform, "sampling_rate": 16000}. For batches and structured metadata, use the model and processor directly as shown below.
This example transcribes two files together, including clips with different durations or sample rates.
language contains a language or dialect label, such as msa, EG, or en. When the model emits [IFM-NoTranscript], postprocessing sets refused=True and returns an empty transcript. Uncomment either prompt option above to request diacritics or code-switching; both can be enabled together.
[!TIP] Pass each recording's original sample rate; the processor resamples it to 16 kHz automatically. Increase
max_new_tokensfor longer recordings: a result withtruncated=Truemeans generation ended before the transcript was complete.
Jais-ASR ships with a vLLM plugin in vllm_plugin/. Install it in its own environment, then start the server:
Transcribe a file through the OpenAI-compatible endpoint:
# 1. Build llama.cpp with Jais-ASR support (once)git clone -b model/K2Horizon https://github.com/MBZUAI-IFM/llama.cpp.git && cd llama.cppgit checkout 35999d101cf2233fc54f09c3c8d599da7303ce02hf download IFM/Jais-ASR-GGUF llama.cpp-jais-asr.patch --local-dir . && git apply llama.cpp-jais-asr.patchcmake -B build -DGGML_CUDA=ON && cmake --build build -j # AMD: -DGGML_HIP=ON, CPU: no flag
# 2. Download one configuration (here 4-bit Medium)hf download IFM/Jais-ASR-GGUF Jais-ASR-Q4_K_M.gguf mmproj-Jais-ASR-Q6_K.gguf --local-dir .
# 3. Start the server, wait until it is ready, then transcribe (add -F language=EG to pin the language or dialect)./build/bin/llama-server -m Jais-ASR-Q4_K_M.gguf --mmproj mmproj-Jais-ASR-Q6_K.gguf --port 8080 &until curl -s localhost:8080/health | grep -q ok; do sleep 1; donecurl localhost:8080/v1/audio/transcriptions -F file=@speech.wavpip install 'transformers==5.16.1' accelerate soundfile soxrimport torchfrom transformers import pipeline
asr = pipeline( "automatic-speech-recognition", model="IFM/Jais-ASR", trust_remote_code=True, device_map="auto", dtype=torch.bfloat16,)result = asr("speech.wav", max_new_tokens=256)print(result["text"])import soundfile as sfimport torchfrom transformers import AutoModelForCausalLM, AutoProcessor
model_id = "IFM/Jais-ASR"processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)processor.tokenizer.padding_side = "left"model = AutoModelForCausalLM.from_pretrained( model_id, trust_remote_code=True, device_map="auto", dtype=torch.bfloat16,).eval()
paths = ["speech_1.wav", "speech_2.wav"]audio = [sf.read(path, dtype="float32") for path in paths]waveforms = [waveform for waveform, _ in audio]sample_rates = [sample_rate for _, sample_rate in audio]
prompt = processor.tokenizer.apply_chat_template( [{"role": "user", "content": [{"type": "audio"}]}], tokenize=False, add_generation_prompt=True, # diacritics=True, # Add Arabic vowel marks. # code_switch=True, # Keep other languages in their original script.)inputs = processor( text=[prompt] * len(waveforms), audio=waveforms, sampling_rate=sample_rates, padding=True, return_tensors="pt",).to(model.device)
with torch.inference_mode(): output = model.generate( **inputs, do_sample=False, max_new_tokens=256, # language="EG", # Set the target language to override automatic detection. )
transcripts = processor.post_process_transcription( output, input_lengths=inputs["input_ids"].shape[1],)for path, result in zip(paths, transcripts): print(path, { "text": result.text, "language": result.language, "refused": result.refused, })pip install 'vllm==0.23.0' 'transformers==5.5.4' accelerate soundfile soxrhf download IFM/Jais-ASR --include "vllm_plugin/*" --local-dir Jais-ASRpip install ./Jais-ASR/vllm_pluginvllm serve IFM/Jais-ASR --trust-remote-code --reasoning-parser jais_asr_tag --max-model-len 2048curl http://localhost:8000/v1/audio/transcriptions \ -F file=@clip.wav \ -F model=IFM/Jais-ASR@misc{jaisasr2026, title = {Introducing Jais-ASR: Arabic Speech Recognition for Every Dialect}, author = {{IFM Team}}, year = {2026}, url = {https://huggingface.co/IFM/Jais-ASR},}