Instructions to use audarai/Audar-ASR-V1-Turbo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use audarai/Audar-ASR-V1-Turbo with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", model="audarai/Audar-ASR-V1-Turbo", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("audarai/Audar-ASR-V1-Turbo", trust_remote_code=True)
model = AutoModelForMultimodalLM.from_pretrained("audarai/Audar-ASR-V1-Turbo", trust_remote_code=True, device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use audarai/Audar-ASR-V1-Turbo with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf audarai/Audar-ASR-V1-Turbo:Q4_K_M

Use Docker

docker model run hf.co/audarai/Audar-ASR-V1-Turbo:Q4_K_M

LM Studio
Jan
Ollama
How to use audarai/Audar-ASR-V1-Turbo with Ollama:
```
ollama run hf.co/audarai/Audar-ASR-V1-Turbo:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use audarai/Audar-ASR-V1-Turbo with Docker Model Runner:
```
docker model run hf.co/audarai/Audar-ASR-V1-Turbo:Q4_K_M
```

Lemonade

How to use audarai/Audar-ASR-V1-Turbo with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull audarai/Audar-ASR-V1-Turbo:Q4_K_M

Run and chat with the model

lemonade run user.Audar-ASR-V1-Turbo-Q4_K_M

List all available models

lemonade list

Atomic Chat

Audar-ASR-V1-Turbo

Commit History

Model card: add benchmark-parity qwen-asr usage (recommended path); mark output-protocol handling as mandatory

371428b
verified

AudarAI-Research commited on 10 days ago

Transformers section: load full bf16 from repo root (from_pretrained + trust_remote_code)

00a507d
verified

Dragonhead commited on 19 days ago

Release full-precision bf16 safetensors at repo root (loads via from_pretrained + trust_remote_code, transformers==4.57.6); remove transformers/ subfolder (remote-code needs root)

c2ebfc4
verified

Dragonhead commited on 19 days ago

Release full-precision bf16 safetensors: add transformers/ section (trust_remote_code, transformers==4.57.6), format badge, nav, tags

0fbb7c0
verified

Dragonhead commited on 19 days ago

Release full-precision weights: bf16 safetensors + Transformers code under transformers/ (Qwen3-ASR arch, trust_remote_code)

e37cf48
verified

Dragonhead commited on 19 days ago

Card: align framing to the technical report (open-weight foundation, adaptation-not-proprietary), add report link

ea262c4
verified

Dragonhead commited on Jul 17

Rebrand custom-code filenames to *_audar_asr.py (auto_map rewired; Qwen3-ASR arch identifiers + Apache-2.0 headers preserved)

cabc19d
verified

Dragonhead commited on Jul 17

Add vLLM (GPU serving) deployment section

ad522e1
verified

Dragonhead commited on Jul 17

Add vLLM-ready W4A16 (4-bit, compressed-tensors) build

764e033
verified

Dragonhead commited on Jul 17

Link GitHub dev hub (AudarAI/Audar-ASR-V1): badge + nav + footer

f16af38
verified

Dragonhead commited on Jul 9

Align alignment stage with paper: GRPO -> KTO (Kahneman-Tversky Optimization) preference alignment

abe3110
verified

Dragonhead commited on Jul 9

Leaderboard table: move Avg WER/CER to first two columns

cb109b1
verified

Dragonhead commited on Jul 8

Update benchmarks: full 35-model Open-AR-ASR leaderboard (Turbo #1) + add cohere-transcribe-arabic-07-2026

cf2bb57
verified

Dragonhead commited on Jul 8

Add measured internal-sample benchmark table (backs the Flash-card comparison claim)

6f8a78a
verified

Dragonhead commited on Jul 7

Precise parameter counts: exact values verified by loading the actual model (thinker.audio_tower / model / lm_head split)

825b6d9
verified

Dragonhead commited on Jul 7

Add missing parameter-count row to Model summary (decoder ~2.0B, full model ~2.35B, verified against actual weights)

efb7c8a
verified

Dragonhead commited on Jul 7

Pre-launch review: fix param/benchmark consistency, align demo sampling, tighten claims

f9f8781
verified

Dragonhead commited on Jul 3

Card: add About AudarAI branding block

e03fcee
verified

Dragonhead commited on Jul 3

GGUF-only distribution: remove safetensors + transformers files (Community tier)

0ff709f
verified

Dragonhead commited on Jul 3

Card v2: proprietary framing + benchmarks + GGUF-only

1bd8507
verified

Dragonhead commited on Jul 3

Add GGUF: Q4_K_M/Q8_0/BF16 decoder + BF16 audio mmproj

7b2a252
verified

Dragonhead commited on Jul 3

Add enterprise model card

dc8b1ff
verified

Dragonhead commited on Jul 3

Fix loadability: add auto_map to config + preprocessor

25257d1
verified

Dragonhead commited on Jul 3

Fix: add lm_head.weight (tied to embed_tokens)

298189d
verified

audar commited on Apr 23

Upload folder using huggingface_hub

eec483a
verified

Dragonhead commited on Apr 23

initial commit

64aad38
verified

Dragonhead commited on Apr 23

Commit History

Model card: add benchmark-parity qwen-asr usage (recommended path); mark output-protocol handling as mandatory 371428b verified

Transformers section: load full bf16 from repo root (from_pretrained + trust_remote_code) 00a507d verified

Release full-precision bf16 safetensors at repo root (loads via from_pretrained + trust_remote_code, transformers==4.57.6); remove transformers/ subfolder (remote-code needs root) c2ebfc4 verified

Release full-precision bf16 safetensors: add transformers/ section (trust_remote_code, transformers==4.57.6), format badge, nav, tags 0fbb7c0 verified

Release full-precision weights: bf16 safetensors + Transformers code under transformers/ (Qwen3-ASR arch, trust_remote_code) e37cf48 verified

Card: align framing to the technical report (open-weight foundation, adaptation-not-proprietary), add report link ea262c4 verified

Rebrand custom-code filenames to *_audar_asr.py (auto_map rewired; Qwen3-ASR arch identifiers + Apache-2.0 headers preserved) cabc19d verified

Add vLLM (GPU serving) deployment section ad522e1 verified

Add vLLM-ready W4A16 (4-bit, compressed-tensors) build 764e033 verified

Link GitHub dev hub (AudarAI/Audar-ASR-V1): badge + nav + footer f16af38 verified

Align alignment stage with paper: GRPO -> KTO (Kahneman-Tversky Optimization) preference alignment abe3110 verified

Leaderboard table: move Avg WER/CER to first two columns cb109b1 verified

Update benchmarks: full 35-model Open-AR-ASR leaderboard (Turbo #1) + add cohere-transcribe-arabic-07-2026 cf2bb57 verified

Add measured internal-sample benchmark table (backs the Flash-card comparison claim) 6f8a78a verified

Precise parameter counts: exact values verified by loading the actual model (thinker.audio_tower / model / lm_head split) 825b6d9 verified

Add missing parameter-count row to Model summary (decoder ~2.0B, full model ~2.35B, verified against actual weights) efb7c8a verified

Pre-launch review: fix param/benchmark consistency, align demo sampling, tighten claims f9f8781 verified

Card: add About AudarAI branding block e03fcee verified

GGUF-only distribution: remove safetensors + transformers files (Community tier) 0ff709f verified

Card v2: proprietary framing + benchmarks + GGUF-only 1bd8507 verified

Add GGUF: Q4_K_M/Q8_0/BF16 decoder + BF16 audio mmproj 7b2a252 verified

Add enterprise model card dc8b1ff verified

Fix loadability: add auto_map to config + preprocessor 25257d1 verified

Fix: add lm_head.weight (tied to embed_tokens) 298189d verified

Upload folder using huggingface_hub eec483a verified

initial commit 64aad38 verified

Model card: add benchmark-parity qwen-asr usage (recommended path); mark output-protocol handling as mandatory

371428b
verified

Transformers section: load full bf16 from repo root (from_pretrained + trust_remote_code)

00a507d
verified

Release full-precision bf16 safetensors at repo root (loads via from_pretrained + trust_remote_code, transformers==4.57.6); remove transformers/ subfolder (remote-code needs root)

c2ebfc4
verified

Release full-precision bf16 safetensors: add transformers/ section (trust_remote_code, transformers==4.57.6), format badge, nav, tags

0fbb7c0
verified

Release full-precision weights: bf16 safetensors + Transformers code under transformers/ (Qwen3-ASR arch, trust_remote_code)

e37cf48
verified

Card: align framing to the technical report (open-weight foundation, adaptation-not-proprietary), add report link

ea262c4
verified

Rebrand custom-code filenames to *_audar_asr.py (auto_map rewired; Qwen3-ASR arch identifiers + Apache-2.0 headers preserved)

cabc19d
verified

Add vLLM (GPU serving) deployment section

ad522e1
verified

Add vLLM-ready W4A16 (4-bit, compressed-tensors) build

764e033
verified

Link GitHub dev hub (AudarAI/Audar-ASR-V1): badge + nav + footer

f16af38
verified

Align alignment stage with paper: GRPO -> KTO (Kahneman-Tversky Optimization) preference alignment

abe3110
verified

Leaderboard table: move Avg WER/CER to first two columns

cb109b1
verified

Update benchmarks: full 35-model Open-AR-ASR leaderboard (Turbo #1) + add cohere-transcribe-arabic-07-2026

cf2bb57
verified

Add measured internal-sample benchmark table (backs the Flash-card comparison claim)

6f8a78a
verified

Precise parameter counts: exact values verified by loading the actual model (thinker.audio_tower / model / lm_head split)

825b6d9
verified

Add missing parameter-count row to Model summary (decoder ~2.0B, full model ~2.35B, verified against actual weights)

efb7c8a
verified

Pre-launch review: fix param/benchmark consistency, align demo sampling, tighten claims

f9f8781
verified

Card: add About AudarAI branding block

e03fcee
verified

GGUF-only distribution: remove safetensors + transformers files (Community tier)

0ff709f
verified

Card v2: proprietary framing + benchmarks + GGUF-only

1bd8507
verified

Add GGUF: Q4_K_M/Q8_0/BF16 decoder + BF16 audio mmproj

7b2a252
verified

Add enterprise model card

dc8b1ff
verified

Fix loadability: add auto_map to config + preprocessor

25257d1
verified

Fix: add lm_head.weight (tied to embed_tokens)

298189d
verified

Upload folder using huggingface_hub

eec483a
verified

initial commit

64aad38
verified