Instructions to use zuu007/gpt2-from-scratch with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use zuu007/gpt2-from-scratch with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="zuu007/gpt2-from-scratch")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("zuu007/gpt2-from-scratch")
model = AutoModelForCausalLM.from_pretrained("zuu007/gpt2-from-scratch", device_map="auto")

Transformers.js

How to use zuu007/gpt2-from-scratch with Transformers.js:

// npm i @huggingface/transformers
import { pipeline } from '@huggingface/transformers';

// Allocate pipeline
const pipe = await pipeline('text-generation', 'zuu007/gpt2-from-scratch');

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use zuu007/gpt2-from-scratch with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zuu007/gpt2-from-scratch"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zuu007/gpt2-from-scratch",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/zuu007/gpt2-from-scratch

SGLang

How to use zuu007/gpt2-from-scratch with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "zuu007/gpt2-from-scratch" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zuu007/gpt2-from-scratch",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "zuu007/gpt2-from-scratch" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zuu007/gpt2-from-scratch",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use zuu007/gpt2-from-scratch with Docker Model Runner:
```
docker model run hf.co/zuu007/gpt2-from-scratch
```

GPT-2 (124M), trained from scratch: base model

The pretrained model: 2.5B tokens of FineWeb-Edu, validation loss 3.300. It continues text; it was not trained to follow instructions. Part of github.com/zuurashad/gpt2-from-scratch, a PyTorch reproduction of GPT-2 small trained from scratch on a single 6GB laptop GPU. Try it in your browser.

The architecture is exactly GPT-2 small: 12 layers, 12 heads, width 768, a 1024-token context and the GPT-2 BPE tokenizer. The weights load into Hugging Face's GPT2LMHeadModel unchanged.

Files

file	what it is
`model.safetensors`	full-precision (fp32) weights
`onnx/model_quantized.onnx`	per-channel int8 ONNX for the browser demo. Validation loss vs fp32: +0.0018 generating token by token (8,192 FineWeb-Edu tokens), +0.0048 over one forward pass (51,200 tokens)
`training.json`	training step, validation loss and the training arguments

Use it

from transformers import pipeline

generate = pipeline("text-generation", model="zuu007/gpt2-from-scratch")
print(generate("Photosynthesis is the process by which", max_new_tokens=60)[0]["generated_text"])

In the browser, with transformers.js: await AutoModelForCausalLM.from_pretrained("zuu007/gpt2-from-scratch", { dtype: "q8" }).

Evaluation

Zero-shot, in fp32, scored by per-choice log-likelihood (acc, and length-normalised acc_norm as in lm-evaluation-harness). The baseline is OpenAI's GPT-2 124M through the same harness.

task	metric	this model	OpenAI GPT-2 124M
hellaswag	acc_norm	27.03%	29.55%
arc_easy	acc_norm	42.30%	38.17%
arc_challenge	acc_norm	23.21%	22.95%
piqa	acc_norm	60.45%	61.81%
openbookqa	acc_norm	27.00%	27.20%
winogrande	acc_norm	52.96%	51.62%
boolq	acc_norm	52.19%	48.64%
lambada	acc	20.80%	32.56%
lambada	perplexity	81.1	18.0

Limitations

This is a 124M-parameter model trained on 2.5B tokens. It writes fluent English, but it is frequently wrong, especially about facts, arithmetic and recent events. Its only alignment is supervised fine-tuning on a small instruction dataset, so it can produce incorrect or inappropriate text.

Data and licences

Pretraining: FineWeb-Edu (ODC-By 1.0)
Fine-tuning (chat model only): databricks-dolly-15k, licensed CC BY-SA 3.0. Treat the chat weights as share-alike.
Code: MIT, derived in part from Andrej Karpathy's build-nanogpt (MIT)

Downloads last month: 16

Safetensors

Model size

0.1B params

Tensor type

F32

zuu007
/

gpt2-from-scratch