rekhta hi2ur model

config.json

@@ -0,0 +1,17 @@
+{
+  "model_type": "custom-transliterator",
+  "architectures": [
+    "CustomTransformerModel"
+  ],
+  "hidden_size": 256,
+  "num_attention_heads": 4,
+  "num_hidden_layers": 3,
+  "dim_feedforward": 512,
+  "max_position_embeddings": 256,
+  "src_vocab_size": 139,
+  "tgt_vocab_size": 141,
+  "pad_token_id": 0,
+  "bos_token_id": 2,
+  "eos_token_id": 3,
+  "unk_token_id": 1
+}

devanagari_bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cebd156b47d0c23766a98e970db86873baf44ada068c7074ed998436bd5e9c7
+size 239237

devanagari_bpe.vocab

@@ -0,0 +1,139 @@
+[PAD]	0
+[UNK]	0
+[CLS]	0
+[SEP]	0
+▁	-1.56837
+ा	-2.72961
+े	-3.09795
+क	-3.12604
+र	-3.14134
+ह	-3.15577
+म	-3.41228
+ी	-3.46207
+त	-3.56156
+न	-3.57827
+स	-3.57865
+़	-3.58176
+ं	-3.72041
+ो	-3.76092
+ि	-3.78206
+ज	-3.89364
+ल	-3.89917
+-	-3.91544
+द	-3.97711
+ब	-4.00613
+्	-4.03211
+ु	-4.11664
+ै	-4.1828
+ग	-4.3143
+ए	-4.42049
+य	-4.42315
+ँ	-4.63556
+ख	-4.65368
+प	-4.67017
+ू	-4.76717
+व	-4.82514
+श	-4.83062
+आ	-4.94678
+अ	-5.0527
+फ	-5.13204
+भ	-5.30514
+च	-5.38654
+इ	-5.45387
+उ	-5.55407
+थ	-5.62956
+ई	-5.6659
+'	-5.66818
+झ	-5.88977
+ड	-5.92759
+ट	-6.11794
+छ	-6.12487
+ौ	-6.2094
+ओ	-6.35333
+औ	-6.65035
+ठ	-6.868
+ध	-6.86917
+ऐ	-7.01473
+घ	-7.14654
+ढ	-7.29567
+ऊ	-8.06043
+,	-9.41882
+!	-9.8114
+.	-10.3114
+‘	-10.5707
+ॉ	-10.7803
+’	-10.8678
+?	-10.8859
+ष	-11.0155
+ण	-11.4962
+ृ	-11.5722
+(	-12.4031
+)	-12.4188
+ञ	-12.6159
+ऑ	-12.7605
+:	-13.1119
+2	-13.734
+1	-14.0549
+a	-14.1693
+3	-14.4469
+।	-14.4877
+ऋ	-14.5747
+e	-14.6213
+ॅ	-14.6213
+4	-14.67
+۔	-14.7754
+r	-14.8932
+ङ	-14.9577
+u	-15.0267
+‍	-15.2679
+*	-15.3632
+5	-15.3632
+i	-15.3632
+ٖ	-15.3632
+6	-15.4686
+7	-15.4686
+;	-15.4686
+b	-15.4686
+h	-15.4686
+s	-15.4686
+8	-15.5863
+२	-15.5863
+n	-15.7199
+v	-15.7199
+l	-15.874
+0	-16.0563
+t	-16.0563
+z	-16.0563
+9	-16.2795
+T	-16.2795
+d	-16.2795
+k	-16.2795
+o	-16.2795
+p	-16.2795
+أ	-16.2795
+३	-16.2795
+_	-16.5672
+ˈ	-16.5672
+“	-16.5672
++	-16.9726
+y	-16.9726
+،	-16.9726
+ٍ	-16.9726
+ऩ	-16.9726
+—	-16.9726
+"	-17.6658
+H	-17.6658
+I	-17.6658
+K	-17.6658
+L	-17.6658
+O	-17.6658
+U	-17.6658
+f	-17.6658
+|	-17.6658
+£	-17.6658
+س	-17.6658
+ع	-17.6658
+ِ	-17.6658
+ٓ	-17.6658
+ک	-17.6658
+ः	-17.6658

h2u_2.0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51c6c40da98a2bc82c3a88200d46fdc00014ae92d4e5a0f1cad9b6b3ac3dcd00
+size 48872290

nastaaliq_bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0e70a3a6728c54b310f196f1e3dda86ea92fac6573a0473fc4529c65f05c696
+size 239191

nastaaliq_bpe.vocab

@@ -0,0 +1,141 @@
+[PAD]	0
+[UNK]	0
+[CLS]	0
+[SEP]	0
+▁	-1.43725
+ا	-2.49726
+ی	-2.61728
+ہ	-2.921
+و	-3.05754
+ر	-3.0802
+ک	-3.08283
+ے	-3.12091
+ن	-3.31991
+م	-3.35435
+ت	-3.45322
+ں	-3.52174
+س	-3.63601
+ب	-3.71838
+ھ	-3.7338
+ل	-3.84319
+د	-3.87154
+ج	-4.17633
+گ	-4.346
+پ	-4.47736
+ش	-4.76856
+ئ	-4.78603
+چ	-4.97538
+آ	-5.09129
+ز	-5.11549
+خ	-5.14642
+ح	-5.30285
+ق	-5.304
+ف	-5.32229
+ع	-5.4835
+ٹ	-5.71517
+ص	-5.91753
+ڑ	-5.97114
+ط	-6.11523
+غ	-6.14773
+ظ	-6.60046
+ؔ	-6.76266
+ڈ	-6.90553
+ؤ	-6.92384
+ض	-7.01754
+ذ	-7.02157
+ۂ	-7.43851
+ث	-7.59653
+ٔ	-9.09115
+ٰ	-9.32497
+،	-9.33314
+'	-9.48661
+!	-9.74387
+ً	-10.0546
+ژ	-10.1661
+۔	-10.5828
+؟	-10.8042
+أ	-10.8382
+.	-11.4494
+(	-12.3291
+)	-12.3343
+ۓ	-12.4488
+ۃ	-12.7346
+ي	-12.7899
+ء	-13.0172
+:	-13.0588
+-	-13.6901
+ك	-13.8409
+ؐ	-13.9386
+۲	-14.0468
+َ	-14.0758
+ه	-14.2009
+ّ	-14.2348
+2	-14.4666
+ُ	-14.5576
+۳	-14.5576
+1	-14.6577
+ؑ	-14.6577
+۱	-14.7689
+ؓ	-14.9631
+ٴ	-14.9631
+۴	-15.1172
+*	-15.2995
+3	-15.4049
+,	-15.5227
+۶	-15.5227
+4	-15.6562
+۵	-15.6562
+i	-15.8104
+ِ	-15.8104
+۷	-15.8104
+۸	-15.8104
+ः	-15.8104
+‘	-15.8104
+’	-15.8104
+a	-15.9927
+e	-15.9927
+r	-15.9927
+s	-15.9927
+؛	-15.9927
+ْ	-15.9927
+0	-16.2158
+A	-16.2158
+b	-16.2158
+d	-16.2158
+n	-16.2158
+u	-16.2158
+5	-16.5035
+7	-16.5035
+I	-16.5035
+R	-16.5035
+ٍ	-16.5035
+“	-16.5035
+8	-16.909
+9	-16.909
+C	-16.909
+E	-16.909
+G	-16.909
+L	-16.909
+U	-16.909
+t	-16.909
+y	-16.909
+ة	-16.909
+ٌ	-16.909
+۹	-16.909
++	-17.6021
+6	-17.6021
+?	-17.6021
+B	-17.6021
+M	-17.6021
+O	-17.6021
+T	-17.6021
+Z	-17.6021
+[	-17.6021
+f	-17.6021
+h	-17.6021
+o	-17.6021
+z	-17.6021
+؎	-17.6021
+ٓ	-17.6021
+ٖ	-17.6021
+۰	-17.6021

readme.md

@@ -0,0 +1,165 @@
+---
+license: other
+language:
+- ur
+- hi
+tags:
+- pytorch
+- transliterations
+- urdu
+- hindi
+- RekhtaLabs
+- Sequence2Sequence
+- Transformers
+---
+![Rekhta Lab Logo](https://www.rekhta.org/Content/Images/RekhtaLogo.png)
+#  Hindi to Urdu Transliteration Model (Character-Level)
+
+
+
+This is a lightweight Transformer-based model trained for **character-level transliteration** of **Hindi poetry into Urdu script**. The model is specially tuned for literary and poetic text, making it ideal for applications involving shayari, nazm, or ghazals.
+
+# Live Inference
+https://rekhtalabs.org/demo/transliterate
+
+
+
+##  Model Overview
+| Feature                 | Value                    |
+|-------------------------|--------------------------|
+| **Architecture**        | Transformer (BART-style) |
+| **Tokenizer**           | Character-level          |
+| **Embedding Size**      | 256                      |
+| **Hidden Size**         | 256 (`d_model`)          |
+| **Feedforward Size**    | 512 (`dim_feedforward`)  |
+| **Encoder Layers**      | 3 (`num_layers`)         |
+| **Decoder Layers**      | 3 (`num_layers`)         |
+| **Attention Heads**     | 4 (`nhead`)              |
+| **Max Sequence Length** | 128 (`max_len`)          |
+
+---
+
+
+
+##  Usage
+
+```python
+from huggingface_hub import snapshot_download
+
+path = snapshot_download(
+    repo_id="rekhtalabs/hi-2-ur-translit",
+    local_dir="./hi-2-ur-translit",
+    local_dir_use_symlinks=False
+)
+
+cd hi-2-ur-translit
+```
+
+```python
+pip install -r requirements.txt
+```
+```python
+import torch
+import sentencepiece as spm
+from torch import nn
+from collections import OrderedDict
+
+
+class PositionalEncoding(nn.Module):
+    def __init__(self, d_model, max_len=5000):
+        super().__init__()
+        pe = torch.zeros(max_len, d_model)
+        position = torch.arange(0, max_len).unsqueeze(1)
+        div_term = torch.exp(torch.arange(0, d_model, 2) * (-torch.log(torch.tensor(10000.0)) / d_model))
+        pe[:, 0::2] = torch.sin(position.float() * div_term)
+        pe[:, 1::2] = torch.cos(position.float() * div_term)
+        self.pe = pe.unsqueeze(0)
+
+    def forward(self, x):
+        return x + self.pe[:, :x.size(1)].to(x.device)
+
+
+class Transformer(nn.Module):
+    def __init__(self, src_vocab_size, tgt_vocab_size, d_model=256, nhead=4, num_layers=3, dim_feedforward=512, max_len=128):
+        super().__init__()
+        self.src_tok_emb = nn.Embedding(src_vocab_size, d_model)
+        self.tgt_tok_emb = nn.Embedding(tgt_vocab_size, d_model)
+        self.pos_encoder = PositionalEncoding(d_model, max_len)
+        self.transformer = nn.Transformer(
+            d_model=d_model,
+            nhead=nhead,
+            num_encoder_layers=num_layers,
+            num_decoder_layers=num_layers,
+            dim_feedforward=dim_feedforward,
+            batch_first=True
+        )
+        self.out = nn.Linear(d_model, tgt_vocab_size)
+
+    def forward(self, src, tgt):
+        src = self.pos_encoder(self.src_tok_emb(src))
+        tgt = self.pos_encoder(self.tgt_tok_emb(tgt))
+        tgt_mask = nn.Transformer.generate_square_subsequent_mask(tgt.size(1)).to(src.device)
+        out = self.transformer(src, tgt, tgt_mask=tgt_mask)
+        return self.out(out)
+
+
+sp_nastaaliq = spm.SentencePieceProcessor(model_file='nastaaliq_bpe.model')
+sp_devanagari = spm.SentencePieceProcessor(model_file='devanagari_bpe.model')
+
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model = Transformer(
+    src_vocab_size=sp_devanagari.get_piece_size(),
+    tgt_vocab_size=sp_nastaaliq.get_piece_size()
+).to(device)
+
+
+checkpoint = torch.load("h2u_2.0.pt", map_location=device)
+state_dict = checkpoint["model_state_dict"]
+new_state_dict = OrderedDict()
+for k, v in state_dict.items():
+    new_k = k.replace("module.", "")
+    new_state_dict[new_k] = v
+model.load_state_dict(new_state_dict)
+model.eval()
+
+
+def transliterate_urdu_to_hindi(text_urdu, max_len=128):
+    src_ids = [2] + sp_devanagari.encode(text_urdu)[:max_len - 2] + [3] 
+    src_tensor = torch.tensor(src_ids).unsqueeze(0).to(device)
+
+    tgt_ids = [2]  # BOS token
+    for _ in range(max_len):
+        tgt_tensor = torch.tensor(tgt_ids).unsqueeze(0).to(device)
+        with torch.no_grad():
+            output = model(src_tensor, tgt_tensor)
+            next_token_logits = output[0, -1, :]
+            next_token_id = torch.argmax(next_token_logits).item()
+
+        if next_token_id == 3:  
+            break
+        tgt_ids.append(next_token_id)
+
+    return sp_nastaaliq.decode(tgt_ids[1:])  
+
+
+res=transliterate_urdu_to_hindi("थम गए हों बहते बहते चम्पई रुख़्सार पर")
+print(res)
+
+
+
+```
+##  Output
+```python
+تھم گئے ہوں بہتے بہتے چمپئی رخسار پر
+```
+
+---
+
+##  Dataset
+
+- Trained on approximately **1300,000 Hindi-Urdu Ghazal and Nazm Pairs**
+- Sourced and curated for transliteration.
+- Character-level alignment ensured for quality
+
+---

requirements.txt

@@ -0,0 +1,25 @@
+filelock==3.18.0
+fsspec==2025.7.0
+Jinja2==3.1.6
+MarkupSafe==3.0.2
+mpmath==1.3.0
+networkx==3.4.2
+nvidia-cublas-cu12==12.6.4.1
+nvidia-cuda-cupti-cu12==12.6.80
+nvidia-cuda-nvrtc-cu12==12.6.77
+nvidia-cuda-runtime-cu12==12.6.77
+nvidia-cudnn-cu12==9.5.1.17
+nvidia-cufft-cu12==11.3.0.4
+nvidia-cufile-cu12==1.11.1.6
+nvidia-curand-cu12==10.3.7.77
+nvidia-cusolver-cu12==11.7.1.2
+nvidia-cusparse-cu12==12.5.4.2
+nvidia-cusparselt-cu12==0.6.3
+nvidia-nccl-cu12==2.26.2
+nvidia-nvjitlink-cu12==12.6.85
+nvidia-nvtx-cu12==12.6.77
+sentencepiece==0.2.0
+sympy==1.14.0
+torch==2.7.1
+triton==3.3.1
+typing_extensions==4.14.1