Skip to content
Audio

jssaluja/indicconformer-stream-72-23

CTC-ASR для Gurbani на пенджабском

1 min read

Это CTC-файнтюн претрейненного энкодера AI4Bharat IndicConformer для пенджабского (pan) с новым 512-piece Gurbani SentencePiece (BPE) словарём и CTC-головой. Модель обучена на приватной речи Punjabi Gurbani и предназначена для распознавания паатха и киртана из Sri Guru Granth Sahib. В сети 115 392 018 параметров.

Ключевые особенности

  • Архитектура: IndicConformer encoder + новая 1×1 convolution-голова, CTC loss.
  • Новый 512-piece Gurbani SentencePiece (BPE) словарь; blank id — 512.
  • Дообучение end-to-end на 183 161 строке; финальные веса — среднее последних 5 эпох.
  • На eval: WER 0.5651, CER 0.2349, NCER 0.1939.
  • Не transformers-checkpoint: pipeline() не загрузит; выход — сырой акустический текст.

Пример

import torch, sentencepiece as spm
from huggingface_hub import hf_hub_download
from nemo.collections.asr.models import ASRModel
repo = 'jssaluja/indicconformer-stream-72-23-sggs-ncer-train-183161-epochs-21-test-2200'
state = torch.load(hf_hub_download(repo, 'final/model.pt'), map_location='cpu')
sp = spm.SentencePieceProcessor(model_file=hf_hub_download(repo, 'final/gurbani_512.model'))
base = ASRModel.restore_from(hf_hub_download('ai4bharat/indicconformer_stt_pa_hybrid_ctc_rnnt_large', 'indicconformer_stt_pa_hybrid_rnnt_large.nemo'), map_location='cpu')
base.encoder.load_state_dict(state['encoder'])
head = torch.nn.Conv1d(base.cfg.encoder.d_model, 513, 1)
ПодробнееWER 0.5651, CER 0.2349, NCER 0.1939 на eval.
Только kirtan (631 строка): WER 0.6160, CER 0.2981, NCER 0.2876.
115 392 018 параметров; не transformers-checkpoint, pipeline() не загрузит.
Для CPU-инференса есть int8 ONNX-экспорт jssaluja/sggs-asr-indic-v0-onnx.
Выход — сырой акустический текст, его нужно сверять с реальным Gurbani.
Discuss Telegram 0 237

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.3 GB · ready files1.9 GB
  • FP8 / INT8recommendedweights 0.6 GB1.2 GB
  • 4-bitweights 0.4 GB0.9 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

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.

    # jssaluja/indicconformer-stream-72-23-sggs-ncer-train-183161-epochs-21-test-2200
    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 jssaluja/indicconformer-stream-72-23-sggs-ncer-train-183161-epochs-21-test-2200 --local-dir ~/models/indicconformer-stream-72-23-sggs-ncer-train-183161-epochs-21-test-2200
    # 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 indicconformer_stt_pa_hybrid_ctc_rnnt_large (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face
Fine-tunes 4All on Hugging Face

About the model

jssaluja/indicconformer-stream-72-23-sggs-ncer-train-183161-epochs-21-test-2200
Author
jssaluja
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

2370

Ask AI about this model

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

Comments