run.py

"""
MANAS - Model for Awadhi Natural Autoregressive Sequences
Interactive inference script. Run: python run.py

Requirements:
    pip install -r requirements.txt

Then download the model:
    python -c "from huggingface_hub import hf_hub_download; hf_hub_download('JayF14/MANAS', 'tulsidas_model.pth', local_dir='.')"
"""

import torch
import torch.nn as nn
from torch.nn import functional as F
import json
import re
import sys
import os

try:
    from indic_transliteration import sanscript
    from indic_transliteration.sanscript import transliterate
    TRANSLITERATION_AVAILABLE = True
except ImportError:
    TRANSLITERATION_AVAILABLE = False
    print("Note: indic-transliteration not installed. Only Devanagari input will work.")
    print("Install with: pip install indic-transliteration\n")

# Hyperparameters (must match training exactly)
block_size = 256
n_embd = 384
n_head = 6
n_layer = 6
dropout = 0.0  # 0 for inference
device = 'cuda' if torch.cuda.is_available() else 'cpu'

# Load vocabulary from vocab.json
VOCAB_FILE = os.path.join(os.path.dirname(__file__), 'vocab.json')
MODEL_FILE = os.path.join(os.path.dirname(__file__), 'tulsidas_model.pth')

if not os.path.exists(VOCAB_FILE):
    print("Error: vocab.json not found. Make sure it is in the same folder as this script.")
    sys.exit(1)

if not os.path.exists(MODEL_FILE):
    print("Error: tulsidas_model.pth not found.")
    print("Download it with:")
    print('  python -c "from huggingface_hub import hf_hub_download; hf_hub_download(\'JayF14/MANAS\', \'tulsidas_model.pth\', local_dir=\'.\')"\n')
    sys.exit(1)

with open(VOCAB_FILE, 'r', encoding='utf-8') as f:
    vocab = json.load(f)

vocab_size = vocab['vocab_size']
stoi = vocab['stoi']
itos = {int(k): v for k, v in vocab['itos'].items()}

def encode(s):
    unknown = [c for c in s if c not in stoi]
    if unknown:
        raise ValueError(f"Character(s) not in model vocabulary: {unknown}")
    return [stoi[c] for c in s]

def decode(l):
    return ''.join([itos[i] for i in l])


# Model architecture
class Head(nn.Module):
    def __init__(self, head_size):
        super().__init__()
        self.key   = nn.Linear(n_embd, head_size, bias=False)
        self.query = nn.Linear(n_embd, head_size, bias=False)
        self.value = nn.Linear(n_embd, head_size, bias=False)
        self.register_buffer('tril', torch.tril(torch.ones(block_size, block_size)))
        self.dropout = nn.Dropout(dropout)
    def forward(self, x):
        B, T, C = x.shape
        k, q = self.key(x), self.query(x)
        wei = q @ k.transpose(-2, -1) * (k.shape[-1] ** -0.5)
        wei = wei.masked_fill(self.tril[:T, :T] == 0, float('-inf'))
        wei = F.softmax(wei, dim=-1)
        return self.dropout(wei) @ self.value(x)

class MultiHeadAttention(nn.Module):
    def __init__(self, num_heads, head_size):
        super().__init__()
        self.heads   = nn.ModuleList([Head(head_size) for _ in range(num_heads)])
        self.proj    = nn.Linear(n_embd, n_embd)
        self.dropout = nn.Dropout(dropout)
    def forward(self, x):
        return self.dropout(self.proj(torch.cat([h(x) for h in self.heads], dim=-1)))

class FeedForward(nn.Module):
    def __init__(self, n_embd):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(n_embd, 4 * n_embd), nn.ReLU(),
            nn.Linear(4 * n_embd, n_embd), nn.Dropout(dropout),
        )
    def forward(self, x): return self.net(x)

class Block(nn.Module):
    def __init__(self, n_embd, n_head):
        super().__init__()
        head_size = n_embd // n_head
        self.sa   = MultiHeadAttention(n_head, head_size)
        self.ffwd = FeedForward(n_embd)
        self.ln1  = nn.LayerNorm(n_embd)
        self.ln2  = nn.LayerNorm(n_embd)
    def forward(self, x):
        x = x + self.sa(self.ln1(x))
        x = x + self.ffwd(self.ln2(x))
        return x

class LanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.token_embedding_table    = nn.Embedding(vocab_size, n_embd)
        self.position_embedding_table = nn.Embedding(block_size, n_embd)
        self.blocks  = nn.Sequential(*[Block(n_embd, n_head=n_head) for _ in range(n_layer)])
        self.ln_f    = nn.LayerNorm(n_embd)
        self.lm_head = nn.Linear(n_embd, vocab_size)

    def forward(self, idx, targets=None):
        B, T = idx.shape
        x = self.token_embedding_table(idx) + self.position_embedding_table(torch.arange(T, device=device))
        logits = self.lm_head(self.ln_f(self.blocks(x)))
        return logits, None

    def generate(self, idx, max_new_tokens):
        for _ in range(max_new_tokens):
            logits, _ = self(idx[:, -block_size:])
            idx_next = torch.multinomial(F.softmax(logits[:, -1, :], dim=-1), num_samples=1)
            idx = torch.cat((idx, idx_next), dim=1)
        return idx


# Load model
print(f"\nUsing device: {device}")
print("Loading MANAS model...")
model = LanguageModel()
model.load_state_dict(torch.load(MODEL_FILE, map_location=device))
model.to(device)
model.eval()
print("Model loaded successfully!\n")

# Common ITRANS overrides for accurate Hindi transliteration
OVERRIDES = {
    "shri ram": "श्री राम",
    "shri rama": "श्री राम",
    "jai shri ram": "जय श्री राम",
    "ram": "राम",
    "sita": "सीता",
    "hanuman": "हनुमान",
    "tulsidas": "तुलसीदास",
    "ramcharitmanas": "रामचरितमानस",
}

def to_devanagari(user_input):
    lower = user_input.lower().strip()
    if lower in OVERRIDES:
        return OVERRIDES[lower], True
    if re.search(r'[a-zA-Z]', user_input) and TRANSLITERATION_AVAILABLE:
        return transliterate(user_input, sanscript.ITRANS, sanscript.DEVANAGARI), True
    return user_input, False


# Interactive loop
print("=" * 55)
print("  MANAS - Model for Awadhi Natural Autoregressive Sequences")
print("=" * 55)
print("Type a Hindi/Awadhi prompt in Devanagari or English.")
print("Examples:  shri ram   |   tulsidas   |   jai shri ram")
print("Type 'quit' or press Ctrl+C to exit.")
print()
print("-" * 55)
print("  IF HINDI TEXT LOOKS BROKEN (overlapping/garbled):")
print()
print("  Windows Terminal (recommended fix):")
print("    Settings > Profiles > Appearance > Font face")
print("    -> Set to:  Nirmala UI   or   Mangal")
print()
print("  VS Code terminal:")
print("    Settings > terminal.integrated.fontFamily")
print("    -> Set to:  Nirmala UI")
print()
print("  Old cmd.exe / PowerShell window:")
print("    Right-click title bar > Properties > Font")
print("    -> Select:  NSimSun  (best fallback)")
print()
print("  Best overall: Use Windows Terminal (free from")
print("  Microsoft Store) with Nirmala UI font.")
print("-" * 55)
print()


while True:
    try:
        user_input = input("You: ").strip()
        if not user_input:
            continue
        if user_input.lower() in ('quit', 'exit', 'q'):
            print("जय श्री राम 🙏")
            break

        hindi_input, was_translated = to_devanagari(user_input)

        if was_translated:
            print(f"[Transliterated] → {hindi_input}")

        try:
            context = torch.tensor([encode(hindi_input)], dtype=torch.long, device=device)
            tokens = model.generate(context, max_new_tokens=400)[0].tolist()
            print(f"\nMANAS:\n{decode(tokens)}\n")
            print("-" * 55)
        except ValueError as e:
            print(f"[Error] {e}")
            print("Tip: Type in Devanagari or standard ITRANS (e.g. 'shri ram').\n")

    except KeyboardInterrupt:
        print("\nजय श्री राम 🙏")
        break