Character-level GPT Model
This is a custom character-level GPT model trained on a text dataset (e.g., Shakespeare). It's a minimal implementation designed for educational purposes.
Model Architecture
The model is a Transformer-based decoder-only architecture, similar to GPT-2, but operating at the character level.
block_size: 1024vocab_size: Dynamically determined from training datan_layer: 12n_head: 12n_embd: 768
How to Use
To use this model, you'll need the pytorch_model.bin (weights) and vocab.json (character mappings).
import torch
import json
from dataclasses import dataclass
import torch.nn as nn
from torch.nn import functional as F
import math
# --- Define your model classes (GPTConfig, CausalSelfAttention, MLP, Block, GPT) here ---
# Copy the relevant classes from your training script.
@dataclass
class GPTConfig:
block_size: int = 1024
vocab_size: int = 50257
n_layer: int = 12
n_head: int = 12
n_embd: int = 768
# ... (CausalSelfAttention, MLP, Block, GPT class definitions) ...
class CausalSelfAttention(nn.Module):
'''A minimal Causal Self-Attention block.'''
def __init__(self, config):
super().__init__()
assert config.n_embd % config.n_head == 0
self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd)
self.c_proj = nn.Linear(config.n_embd, config.n_embd)
self.c_proj.NANGPT_SCALE_INIT = 1.0 / math.sqrt(2.0 * config.n_layer)
self.n_head = config.n_head
self.n_embd = config.n_embd
self.register_buffer("bias", torch.tril(torch.ones(config.block_size, config.block_size))
.view(1, 1, config.block_size, config.block_size))
def forward(self, x):
B, T, C = x.size()
q, k, v = self.c_attn(x).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)
att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf'))
att = F.softmax(att, dim=-1)
y = att @ v
y = y.transpose(1, 2).contiguous().view(B, T, C)
y = self.c_proj(y)
return y
class MLP(nn.Module):
'''A minimal Multi-Layer Perceptron block.'''
def __init__(self, config):
super().__init__()
self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd)
self.gelu = nn.GELU(approximate='tanh')
self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd)
self.c_proj.NANGPT_SCALE_INIT = 1.0 / math.sqrt(2.0 * config.n_layer)
def forward(self, x):
x = self.c_fc(x)
x = self.gelu(x)
x = self.c_proj(x)
return x
class Block(nn.Module):
'''A minimal Transformer Block consisting of Attention and MLP.'''
def __init__(self, config):
super().__init__()
self.ln_1 = nn.LayerNorm(config.n_embd)
self.attn = CausalSelfAttention(config)
self.ln_2 = 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(nn.Module):
'''The full GPT model composed of Blocks.'''
def __init__(self, config):
super().__init__()
self.config = config
self.transformer = nn.ModuleDict(dict(
wte = nn.Embedding(config.vocab_size, config.n_embd),
wpe = nn.Embedding(config.block_size, config.n_embd),
h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
ln_f = nn.LayerNorm(config.n_embd),
))
self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
self.transformer.wte.weight = self.lm_head.weight # Weight tying
self.apply(self._init_weights)
def get_num_params(self, non_embedding=True):
n_params = sum(p.numel() for p in self.parameters())
if non_embedding:
n_params -= self.transformer.wpe.weight.numel()
return n_params
def _init_weights(self, module):
if isinstance(module, nn.Linear):
std = 0.02
if hasattr(module, 'NANGPT_SCALE_INIT'):
std *= module.NANGPT_SCALE_INIT
torch.nn.init.normal_(module.weight, mean=0.0, std=std)
if module.bias is not None:
torch.nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
elif isinstance(module, nn.LayerNorm):
torch.nn.init.zeros_(module.bias)
torch.nn.init.ones_(module.weight)
def forward(self, idx, targets=None):
device = idx.device
B, T = idx.size()
assert T <= self.config.block_size, f"Cannot forward sequence of length {T}, block size is only {self.config.block_size}"
pos = torch.arange(0, T, dtype=torch.long, device=device).unsqueeze(0)
tok_emb = self.transformer.wte(idx)
pos_emb = self.transformer.wpe(pos)
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)
loss = None
if targets is not None:
loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1)
return logits, loss
# --- Custom tokenizer based on vocab.json ---
class SimpleCharTokenizer:
def __init__(self, vocab_file):
with open(vocab_file, 'r') as f:
vocab_data = json.load(f)
self.stoi = vocab_data['stoi']
self.itos = {int(k): v for k, v in vocab_data['itos'].items()} # keys are string in json
self.vocab_size = vocab_data['vocab_size']
def encode(self, s):
return [self.stoi[c] for c in s]
def decode(self, l):
return ''.join([self.itos[i] for i in l])
# --- Generation function (simplified) ---
def generate_from_hf(model, tokenizer, start_str, max_new_tokens, temperature=1.0, top_k=50, device='cpu'):
model.eval()
B, T_model = 1, model.config.block_size # Model's block_size
start_ids = tokenizer.encode(start_str)
x = (torch.tensor(start_ids, dtype=torch.long, device=device)[None, ...])
x = x[:, -T_model:] # Truncate if start string is too long for model's block_size
for _ in range(max_new_tokens):
# crop context if necessary
x_cond = x if x.size(1) <= T_model else x[:, -T_model:]
with torch.no_grad():
logits, _ = model(x_cond)
logits = logits[:, -1, :] / temperature
if top_k is not None:
v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
logits[logits < v[:, [-1]]] = -float('Inf')
probs = F.softmax(logits, dim=-1)
idx_next = torch.multinomial(probs, num_samples=1)
x = torch.cat((x, idx_next), dim=1)
if tokenizer.stoi.get('
') is not None and idx_next.item() == tokenizer.stoi.get('
'):
break
return tokenizer.decode(x[0].tolist())
# Example usage:
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
# # Load config and vocab
# with open('my_gpt_model/config.json', 'r') as f:
# model_config_dict = json.load(f)
# model_config = GPTConfig(**model_config_dict)
#
# tokenizer = SimpleCharTokenizer('my_gpt_model/vocab.json')
# model = GPT(model_config).to(device)
# model.load_state_dict(torch.load('my_gpt_model/pytorch_model.bin', map_location=device))
#
# prompt = "First Citizen:"
# generated_text = generate_from_hf(model, tokenizer, prompt, max_new_tokens=200, temperature=0.9, device=device)
# print(generated_text)
Files in . directory:
pytorch_model.bin: Contains the model's state dictionary (weights).vocab.json: Contains the character-to-integer (stoi) and integer-to-character (itos) mappings.config.json: Contains the model's configuration parameters (GPTConfig).
How to Load and Generate Text
# (Refer to the example usage in the code block above for loading and generating text)
Note: The model architecture classes (GPTConfig, CausalSelfAttention, MLP, Block, GPT) and the generate function itself are part of the model's definition and would need to be present in your environment when loading the model from Hugging Face. The README.md includes these definitions for clarity and ease of use.