Skip to content
Audio

Qwen3-ASR-1.7B-ft-af51-cap900-1106

ASR-файнтюн для 51 африканского языка

1 min readupdated

Это дообученная версия Qwen/Qwen3-ASR-1.7B (2.04B параметров по метаданным) для automatic-speech-recognition. Файнтюн выполнен на 51 языке с Sunbird speech-to-text pipeline; в карточке указан run Qwen3-ASR-1.7B-full-lr2e-05-ep5-af51-1106-c8a8. Модель принимает аудио и возвращает текст, а при неизвестном языке сначала называет его.

Ключевые особенности

  • В списке языков есть Luganda, Swahili, Yoruba, Zulu, Hausa, Amharic, Afrikaans, English и French.
  • Для лучшего результата язык передаётся явно через language=; без него модель сначала определяет язык.
  • Работает через пакет qwen-asr==0.0.6; нужен transformers 4.57.6.
  • Лучший чекпоинт — checkpoint-1640, eval_WER_mean = 0.4334.
  • Для декодирования как при обучении нужен FlashAttention-2 и attn_implementation="flash_attention_2".

Пример

import torch
from qwen_asr import Qwen3ASRModel
from qwen_asr.inference.utils import SUPPORTED_LANGUAGES
SUPPORTED_LANGUAGES.append('Luganda')
model = Qwen3ASRModel.from_pretrained('huwenjie333/Qwen3-ASR-1.7B-ft-af51-cap900-1106', dtype=torch.bfloat16, device_map='cuda:0')
print(model.transcribe(audio='audio.wav', language='Luganda')[0].text)
ПодробнееЛучший чекпоинт: checkpoint-1640, eval_WER_mean = 0.4334.
Установка: qwen-asr==0.0.6, transformers 4.57.6.
Для декодирования как при обучении нужен FlashAttention-2 и attn_implementation="flash_attention_2".
Без указания языка модель сначала определяет его сама.
Discuss Telegram 0 17

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # huwenjie333/Qwen3-ASR-1.7B-ft-af51-cap900-1106
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download huwenjie333/Qwen3-ASR-1.7B-ft-af51-cap900-1106 --local-dir ~/models/Qwen3-ASR-1.7B-ft-af51-cap900-1106
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Qwen3-ASR-1.7B-ft-af51-cap900-1106 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

huwenjie333/Qwen3-ASR-1.7B-ft-af51-cap900-1106
Author
huwenjie333
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

170

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Robotics

LIME

LIME