---
tags:
  - knowledge-editing
  - circuit-entropy
  - gpt2-xl
license: mit
---

# gpt2-xl — FT-vanilla-4 Single-Fact Edit

Model edited with **FT-vanilla-4** (Circuit Entropy Regularization for Knowledge Editing).

## Edit

| | |
|---|---|
| **Prompt** | `The Eiffel Tower is located in the city of` |
| **Target** | `Berlin` |
| **Method** | FT-vanilla-4 |
| **Lambda** | 0.0 |
| **Edit success** | False |

## Training Config

| Parameter | Value |
|---|---|
| Steps | 20 |
| Learning rate | 1e-05 |
| Weight decay | 0.01 |
| Grad clip | 1.0 |
| Lambda (entropy) | 0.0 |
| EAP-IG steps | 5 |
| Seed | 42 |

## Final Metrics

| Metric | Value |
|---|---|
| Final L_CE | 11.296534 |
| Final KL | -0.000000 |
| Final H(C) | 9.3054 |
| Final delta_H | 0.0000 |

## Usage

```python
from transformer_lens import HookedTransformer
import torch

model = HookedTransformer.from_pretrained("gpt2-xl")
state_dict = torch.load("model_state_dict.pt", map_location="cpu")
model.load_state_dict(state_dict)

tokens = model.to_tokens("The Eiffel Tower is located in the city of")
out = model.generate(tokens, max_new_tokens=10, do_sample=False)
print(model.tokenizer.decode(out[0]))
```

## Paper

Circuit Entropy Regularization for Knowledge Editing (NeurIPS 2026 submission)