Architectures / NLP/LLM
🧠 GPT-2
GPT-2 Small — causal transformer block (768D, 12 heads, 4× FFN)
Every number on this page is computed from the graph by the same functions the app runs, not written by hand.
When to pick it
Structure
12 layers. Output shapes are propagated from the input shape, batch dimension excluded.
| Layer | Type | Parameters | Output shape | |
|---|---|---|---|---|
| 1 | tokens | Input | shape=[1, 1024] | 1 × 1024 |
| 2 | token_embed | Embedding | 1 × 1024 × 768 | |
| 3 | pos_embed | Positional Encoding | embedDim=768, maxLen=1024 | 1 × 1024 × 768 |
| 4 | ln_1 | LayerNorm | normalizedShape=768 | 1 × 1024 × 768 |
| 5 | attn | Causal Attention | embedDim=768, numHeads=12 | 1 × 1024 × 768 |
| 6 | residual_1 | Add | 1 × 1024 × 768 | |
| 7 | ln_2 | LayerNorm | normalizedShape=768 | 1 × 1024 × 768 |
| 8 | mlp | Feed Forward | ffDim=3072 | 1 × 1024 × 768 |
| 9 | residual_2 | Add | 1 × 1024 × 768 | |
| 10 | ln_f | LayerNorm | normalizedShape=768 | 1 × 1024 × 768 |
| 11 | lm_head | Linear | outFeatures=50257 | 1 × 1024 × 50257 |
| 12 | logits | Output | 1 × 1024 × 50257 |
What the verifier says
The same 41 structural checks that run on every edit in the app, on this graph.
No finding. Shapes propagate end to end, every divisibility condition holds, and no advisory rule fires. See the checks.
The PyTorch it exports
Generated from the graph above. First 46 lines; the app exports the whole file, plus the training loop, the data contract and a deploy bundle.
# Architecture designed with Neurarch: https://neurarch.com
# PyTorch: compatible with Python 3.8+ and torch>=1.12
# Colab: pip install torch torchvision (usually pre-installed)
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Tuple
class GPT_2Block(nn.Module):
def __init__(self):
super().__init__()
self.embedding_1 = nn.Embedding(50257, 768)
self.layerNorm_1 = nn.LayerNorm(768)
self.causalAttention_1 = nn.MultiheadAttention(embed_dim=768, num_heads=12, batch_first=True)
self.layerNorm_2 = nn.LayerNorm(768)
self.feedForward_1 = nn.Sequential(
nn.Linear(768, 3072),
nn.ReLU(),
nn.Linear(3072, 768)
)
self.layerNorm_3 = nn.LayerNorm(768)
self.linear_1 = nn.Linear(768, 50257)
def forward(self, x):
# tokens shape: [1,1024]
embedding_ding_1 = self.embedding_1(x)
# positionalEncoding: add positional encoding externally (e.g. sinusoidal or learned PE)
layer_norm_Norm_1 = self.layerNorm_1(embedding_ding_1)
causal_attention_tion_1 = self.causalAttention_1(layer_norm_Norm_1, layer_norm_Norm_1, layer_norm_Norm_1)[0]
add_add_1 = causal_attention_tion_1 + embedding_ding_1
layer_norm_Norm_2 = self.layerNorm_2(add_add_1)
feed_forward_ward_1 = self.feedForward_1(layer_norm_Norm_2)
add_add_2 = feed_forward_ward_1 + add_add_1
layer_norm_Norm_3 = self.layerNorm_3(add_add_2)
linear_near_1 = self.linear_1(layer_norm_Norm_3)
# Output
return linear_near_1
if __name__ == '__main__':
model = GPT_2Block()
model.eval()
x = torch.randint(0, 50000, (1, 1024)) # (batch, features)For agents
This architecture is machine-readable end to end. An agent can list the set, fetch this graph, edit it, and have the edit verified before any GPU time is spent.