YAML Metadata Warning:The pipeline tag "text2text-generation" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
Hinglish → Devanagari Transliteration Model (v3)
15M-param character-level seq2seq Transformer for romanized Hindi (Hinglish)
→ Devanagari transliteration, built for a Chinese-smart-pinyin-style Hindi IME.
Trained on IB-Emper/hinglish-translit-v2
(548,943 pairs) plus the Dakshina lexicon and a capped Aksharantar split.
Files
best.pt— fp32 checkpoint, ~60MBbest_int8.pt— dynamically INT8-quantized checkpoint (feedforward layers only; attention projections stay fp32 — PyTorch's dynamic quantization does not touchnn.MultiheadAttentionprojections), ~37MBmodel.py— architecture + tokenizer + checkpoint load/save (needed to load either file)
Architecture
d_model=384, nhead=6, 4 encoder + 4 decoder layers, dim_feedforward=1280, dropout=0.1, max_len=64, norm_first (pre-LN), weight-tied embedding/output. Char-level tokenizer, frozen vocab stored in the checkpoint. 15.13M params.
Usage
from model import load_checkpoint, BOS, EOS
import torch
model, tok, extra = load_checkpoint("best.pt", map_location="cpu")
# beam search / decoding left to the caller -- see the project repo for a
# reference implementation (KV-cached beam search decoder).
Loading best_int8.pt requires reconstructing the model, applying
torch.quantization.quantize_dynamic(model, {torch.nn.Linear}, dtype=torch.qint8),
then calling load_state_dict on the result — the packed quantized weights
won't load directly into a plain fp32 model.
Eval (Dakshina lexicon, Hindi test set, beam+LM rescoring)
| acc@1 | acc@3 | |
|---|---|---|
| ALL | 62.5% | 77.0% |
| FREQ tier (own corpus-frequency tiering) | 84.2% | — |
For comparison, IndicXlit (AI4Bharat/Microsoft, 11M params split across 21 languages) reports 60.5% acc@1 / 81.9% acc@3 unreranked, 72.4% acc@1 with their own reranking, on the same official Dakshina test set. This model is competitive on acc@1, not superior — it loses to IndicXlit's reranked top-1 and to their top-3 even before reranking. It was built and tuned for casual/ code-switched chat-style input specifically, which IndicXlit was not trained or evaluated on; that comparison has not been run head-to-head.
Known limitations
- Confusable near-homophone pairs (मेरा/मारा, क्या/किया-type errors) may or may not always have a correct prediction.
- ~13% of training pairs retain inline Latin code-switch words on the Devanagari side by design (realistic chat register), not a bug.
- Not yet quantized beyond dynamic INT8;