inference_cuda.py

#!/usr/bin/env python3
"""
CUDA-optimized inference script for Ursa Minor Smashed model
Requires CUDA-capable GPU
"""

import torch
import torch.nn.functional as F
import argparse
import tiktoken
from typing import Optional, List, Tuple
import warnings
warnings.filterwarnings('ignore')

# Direct PyTorch Implementation
class GPTConfig:
    def __init__(self, **kwargs):
        self.block_size = kwargs.get('block_size', 1024)
        self.vocab_size = kwargs.get('vocab_size', 50304)
        self.n_layer = kwargs.get('n_layer', 12)
        self.n_head = kwargs.get('n_head', 12)
        self.n_embd = kwargs.get('n_embd', 768)

class CausalSelfAttention(torch.nn.Module):
    def __init__(self, config):
        super().__init__()
        assert config.n_embd % config.n_head == 0
        self.c_attn = torch.nn.Linear(config.n_embd, 3 * config.n_embd)
        self.c_proj = torch.nn.Linear(config.n_embd, config.n_embd)
        self.n_head = config.n_head
        self.n_embd = config.n_embd

    def forward(self, x):
        B, T, C = x.size()
        qkv = self.c_attn(x)
        q, k, v = qkv.split(self.n_embd, dim=2)
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
        y = F.scaled_dot_product_attention(q, k, v, is_causal=True)
        y = y.transpose(1, 2).contiguous().view(B, T, C)
        y = self.c_proj(y)
        return y

class MLP(torch.nn.Module):
    def __init__(self, config):
        super().__init__()
        self.c_fc = torch.nn.Linear(config.n_embd, 4 * config.n_embd)
        self.gelu = torch.nn.GELU(approximate='tanh')
        self.c_proj = torch.nn.Linear(4 * config.n_embd, config.n_embd)

    def forward(self, x):
        x = self.c_fc(x)
        x = self.gelu(x)
        x = self.c_proj(x)
        return x

class Block(torch.nn.Module):
    def __init__(self, config):
        super().__init__()
        self.ln_1 = torch.nn.LayerNorm(config.n_embd)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = torch.nn.LayerNorm(config.n_embd)
        self.mlp = MLP(config)

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x

class GPT(torch.nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        
        self.transformer = torch.nn.ModuleDict(dict(
            wte = torch.nn.Embedding(config.vocab_size, config.n_embd),
            wpe = torch.nn.Embedding(config.block_size, config.n_embd),
            h = torch.nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = torch.nn.LayerNorm(config.n_embd),
        ))
        self.lm_head = torch.nn.Linear(config.n_embd, config.vocab_size, bias=False)
        
        # Weight tying
        self.transformer.wte.weight = self.lm_head.weight

    def forward(self, idx):
        B, T = idx.size()
        assert T <= self.config.block_size, f"Sequence length {T} exceeds block size {self.config.block_size}"
        
        pos = torch.arange(0, T, dtype=torch.long, device=idx.device)
        pos_emb = self.transformer.wpe(pos)
        tok_emb = self.transformer.wte(idx)
        x = tok_emb + pos_emb
        
        for block in self.transformer.h:
            x = block(x)
            
        x = self.transformer.ln_f(x)
        logits = self.lm_head(x)
        
        return logits

def apply_repetition_penalty(logits: torch.Tensor, token_ids: List[int], penalty: float = 1.1):
    """Apply repetition penalty to logits"""
    for token_id in set(token_ids):
        logits[0, token_id] /= penalty
    return logits

def top_k_top_p_filtering(logits: torch.Tensor, top_k: int = 50, top_p: float = 0.9):
    """Filter logits using top-k and/or top-p (nucleus) filtering"""
    if top_k > 0:
        values, indices = torch.topk(logits, min(top_k, logits.size(-1)))
        logits[logits < values[:, [-1]]] = float('-inf')
    
    if top_p < 1.0:
        sorted_logits, sorted_indices = torch.sort(logits, descending=True)
        cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
        
        # Remove tokens with cumulative probability above the threshold
        sorted_indices_to_remove = cumulative_probs > top_p
        sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
        sorted_indices_to_remove[..., 0] = 0
        
        indices_to_remove = sorted_indices_to_remove.scatter(1, sorted_indices, sorted_indices_to_remove)
        logits[indices_to_remove] = float('-inf')
    
    return logits

def generate_direct(
    model: GPT,
    prompt: str,
    max_new_tokens: int = 100,
    temperature: float = 0.8,
    top_k: int = 50,
    top_p: float = 0.9,
    repetition_penalty: float = 1.1
):
    """Generate text using CUDA-optimized PyTorch implementation"""
    device = "cuda"
    
    # Initialize tokenizer
    enc = tiktoken.get_encoding("gpt2")
    
    # Encode prompt
    tokens = enc.encode(prompt)
    x = torch.tensor(tokens, dtype=torch.long, device=device).unsqueeze(0)
    
    model.eval()
    generated_tokens = []
    
    with torch.no_grad():
        for _ in range(max_new_tokens):
            # Get logits with CUDA autocast for performance
            with torch.cuda.amp.autocast(dtype=torch.bfloat16):
                logits = model(x)
            
            # Focus on last token
            logits = logits[:, -1, :] / temperature
            
            # Apply repetition penalty
            if repetition_penalty > 1.0 and len(generated_tokens) > 0:
                logits = apply_repetition_penalty(logits, generated_tokens[-20:], repetition_penalty)
            
            # Apply top-k and top-p filtering
            filtered_logits = top_k_top_p_filtering(logits, top_k=top_k, top_p=top_p)
            
            # Sample
            probs = F.softmax(filtered_logits, dim=-1)
            next_token = torch.multinomial(probs, num_samples=1)
            
            # Append to sequence
            x = torch.cat([x, next_token], dim=1)
            generated_tokens.append(next_token.item())
            
            # Stop if EOS token
            if next_token.item() == enc.eot_token:
                break
            
            # Truncate if exceeding block size
            if x.size(1) > model.config.block_size:
                x = x[:, -model.config.block_size:]
    
    # Decode
    all_tokens = tokens + generated_tokens
    return enc.decode(all_tokens)

def load_model_direct(checkpoint_path: str):
    """Load model from a PyTorch checkpoint - CUDA optimized"""
    if not torch.cuda.is_available():
        raise RuntimeError("CUDA is not available. Use inference_cpu.py for CPU inference.")
    
    device = "cuda"
    print(f"Loading model from checkpoint: {checkpoint_path}")
    
    # Create a dummy class to handle train_gpt2.GPTConfig references
    import sys
    import types
    
    # Create a fake train_gpt2 module to handle the reference
    train_gpt2_module = types.ModuleType('train_gpt2')
    
    class DummyGPTConfig:
        def __init__(self, **kwargs):
            for k, v in kwargs.items():
                setattr(self, k, v)
    
    train_gpt2_module.GPTConfig = DummyGPTConfig
    sys.modules['train_gpt2'] = train_gpt2_module
    
    try:
        # Load to CPU first to avoid device issues
        checkpoint = torch.load(checkpoint_path, map_location="cpu", weights_only=False)
    finally:
        # Clean up
        if 'train_gpt2' in sys.modules:
            del sys.modules['train_gpt2']
    
    # Handle the config - it might be a train_gpt2.GPTConfig object
    config_obj = checkpoint['config']
    if hasattr(config_obj, '__dict__'):
        # If it's an object, extract its attributes
        config_dict = vars(config_obj)
    else:
        # If it's already a dict
        config_dict = config_obj
    
    config = GPTConfig(**config_dict)
    model = GPT(config)
    model.load_state_dict(checkpoint['model'])
    model.to(device)
    
    # Enable optimizations
    model = torch.compile(model) if hasattr(torch, 'compile') else model
    
    return model

def main():
    parser = argparse.ArgumentParser(description="Generate text with Ursa Minor Smashed model (CUDA)")
    parser.add_argument("--model", type=str, default="model_optimized.pt",
                        help="Path to model checkpoint (.pt file)")
    parser.add_argument("--prompt", type=str, default="Hello, I'm a language model",
                        help="Input prompt")
    parser.add_argument("--max-tokens", type=int, default=100,
                        help="Maximum number of tokens to generate")
    parser.add_argument("--temperature", type=float, default=0.8,
                        help="Sampling temperature (0.1=conservative, 1.0=creative)")
    parser.add_argument("--top-k", type=int, default=50,
                        help="Top-k sampling (0=disabled)")
    parser.add_argument("--top-p", type=float, default=0.9,
                        help="Top-p (nucleus) sampling")
    parser.add_argument("--repetition-penalty", type=float, default=1.1,
                        help="Repetition penalty (1.0=disabled)")
    
    args = parser.parse_args()
    
    # Load model from checkpoint
    model = load_model_direct(args.model)
    
    result = generate_direct(
        model,
        args.prompt,
        args.max_tokens,
        args.temperature,
        args.top_k,
        args.top_p,
        args.repetition_penalty
    )
    
    print("\nGenerated text:")
    print("-" * 50)
    print(result)
    print("-" * 50)

if __name__ == "__main__":
    main()