app.py

import gradio as gr
import torch
import torch.nn as nn
import time

# Define the custom model class with detailed layer structures
class Head(nn.Module):
    def __init__(self, head_size):
        super().__init__()
        self.key = nn.Linear(64, head_size, bias=False)
        self.query = nn.Linear(64, head_size, bias=False)
        self.value = nn.Linear(64, head_size, bias=False)
        self.register_buffer('tril', torch.tril(torch.ones(32, 32)))
        self.dropout = nn.Dropout(0.1)

    def forward(self, x):
        B, T, C = x.shape
        k = self.key(x)
        q = self.query(x)
        wei = q @ k.transpose(-2, -1) * C**-0.5
        wei = wei.masked_fill(self.tril[:T, :T] == 0, float('-inf'))
        wei = nn.functional.softmax(wei, dim=-1)
        wei = self.dropout(wei)
        v = self.value(x)
        return wei @ v

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(64, 64)
        self.dropout = nn.Dropout(0.1)

    def forward(self, x):
        out = torch.cat([h(x) for h in self.heads], dim=-1)
        return self.dropout(self.proj(out))

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(0.1),
        )

    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 BigramLanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.token_embedding_table = nn.Embedding(61, 64)
        self.position_embedding_table = nn.Embedding(32, 64)
        self.blocks = nn.Sequential(*[Block(64, n_head=4) for _ in range(4)])
        self.ln_f = nn.LayerNorm(64)
        self.lm_head = nn.Linear(64, 61)

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

    def generate(self, idx, max_new_tokens, temperature=0.7):
        for _ in range(max_new_tokens):
            idx_cond = idx[:, -32:]  # Truncate to the latest 32 tokens
            logits, _ = self(idx_cond)
            logits = logits[:, -1, :]  # Get the logits for the last token
            logits = logits / temperature  # Apply temperature control
            probs = nn.functional.softmax(logits, dim=-1)
            idx_next = torch.multinomial(probs, num_samples=1)
            idx_next = torch.clamp(idx_next, min=0, max=60)  # Strictly enforce index range [0, 60]
            idx = torch.cat((idx, idx_next), dim=1)
        return idx

# Load the model with strict=False to handle missing or unexpected keys
def load_model():
    model = BigramLanguageModel()
    model_url = "https://huggingface.co/yoonusajwardapiit/triptuner/resolve/main/pytorch_model.bin"
    model_weights = torch.hub.load_state_dict_from_url(model_url, map_location=torch.device('cpu'), weights_only=True)
    model.load_state_dict(model_weights, strict=False)
    model.eval()
    return model

model = load_model()

# Define a comprehensive character set based on training data
chars = sorted(list(set("abcdefghijklmnopqrstuvwxyz0123456789 .,!?-:;'\"\n")))
stoi = {ch: i for i, ch in enumerate(chars)}
itos = {i: ch for i, ch in enumerate(chars)}
encode = lambda s: [stoi.get(c, stoi.get(c.lower(), -1)) for c in s if c in stoi or c.lower() in stoi]  # Handles both cases
decode = lambda l: ''.join([itos[i] for i in l if i < len(itos)])  # Ensures index is within bounds

# Function to generate text using the model
def generate_text(prompt):
    try:
        start_time = time.time()
        print(f"Received prompt: {prompt}")
        encoded_prompt = encode(prompt)
        
        # Check for out-of-vocabulary indices
        if any(idx == -1 for idx in encoded_prompt):
            return "Error: Input contains characters not in the model vocabulary."

        # Ensure the prompt length fits within the block size
        if len(encoded_prompt) > 32:
            encoded_prompt = encoded_prompt[:32]  # Truncate to fit block size
        
        context = torch.tensor([encoded_prompt], dtype=torch.long)
        print(f"Encoded prompt: {context}")
        
        with torch.no_grad():
            generated = model.generate(context, max_new_tokens=20, temperature=0.7)  # Adjust temperature
            print(f"Generated tensor: {generated}")
        
        result = decode(generated[0].tolist())
        print(f"Decoded result: {result}")

        # Post-process to clean up and make output more readable
        cleaned_result = result.replace('\n', ' ').strip()
        print(f"Cleaned result: {cleaned_result}")
        print(f"Processing time: {time.time() - start_time:.2f}s")
        return cleaned_result
    except Exception as e:
        print(f"Error during generation: {e}")
        return f"Error: {str(e)}"

# Create a Gradio interface
interface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=2, placeholder="Enter a location or prompt..."),
    outputs="text",
    title="Triptuner Model",
    description="Generate itineraries for locations in Sri Lanka's Central Province."
)

# Launch the interface
interface.launch()