Instructions to use jfdasher/gptlab-57m-3e17 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jfdasher/gptlab-57m-3e17 with Transformers:

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

pipe = pipeline("text-generation", model="jfdasher/gptlab-57m-3e17")

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

tokenizer = AutoTokenizer.from_pretrained("jfdasher/gptlab-57m-3e17")
model = AutoModelForCausalLM.from_pretrained("jfdasher/gptlab-57m-3e17", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use jfdasher/gptlab-57m-3e17 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jfdasher/gptlab-57m-3e17"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jfdasher/gptlab-57m-3e17",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/jfdasher/gptlab-57m-3e17

SGLang

How to use jfdasher/gptlab-57m-3e17 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 "jfdasher/gptlab-57m-3e17" \
    --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": "jfdasher/gptlab-57m-3e17",
		"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 "jfdasher/gptlab-57m-3e17" \
        --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": "jfdasher/gptlab-57m-3e17",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use jfdasher/gptlab-57m-3e17 with Docker Model Runner:
```
docker model run hf.co/jfdasher/gptlab-57m-3e17
```

jfdasher/gptlab-57m-3e17

A 57,132,544-parameter Llama-style model trained from scratch in plain PyTorch as part of a compute-optimal scaling-law study (run test/3e+17).

Cost: this run: 0.37 GPU-hours, $1.47


Parameters (total, tied embeddings)	57,132,544
Parameters (non-embedding)	52,938,240
Layers × width	17 × 512
Training tokens	875,167,744
Training compute	3e+17 FLOPs (6·N·D)
Val loss (nats/token)	3.0407
Val bits per byte	1.1332
Tokenizer	own 8192-token byte-level BPE
Context	512 tokens

Training data: FineWeb-Edu sample-10BT (ODC-By), document-hash train/val split. W&B run: https://forge.coreweave.com/wandb/jfdasher-thermopylae-systems/gptlab-scaling/runs/2e93fa9c0ffe · code commit: 0e18694c7f00d5b3f279e693d29b994b091bdbfc-dirty

Samples (fixed prompts, temperature 0.8, top-k 50)

The water cycle is a process where the water is sealed between the seams, but not between the seams. This is called frying. If the water is frying then the seams, and the water-saturated part,
In 1492, the Rwenzori people were allowed to use their land in a way that the Mongols took away from the people. It was the second time when the Naharashtra people had gained land in the
To solve a quadratic equation, the pre-conditioned state of the equation assumes that the equation is perfect. Since the state of the equations is constant, the transaction is always positive and the equation
The most important idea in biology is to “draw” the light of the spontaneous movement. It’s not about the light; it’s about looking for the location or direction of the light. The light is also like a line, the light of

Intended use and limitations

Research and teaching only. Small, English-only, trained on web text with FineWeb-Edu's quality filtering and nothing else: no instruction tuning, no safety tuning. It will produce false and incoherent text.

Downloads last month: -

Safetensors

Model size

57.1M params

Tensor type

F32

jfdasher
/

gptlab-57m-3e17

jfdasher/gptlab-57m-3e17

Samples (fixed prompts, temperature 0.8, top-k 50)

Intended use and limitations

Dataset used to train jfdasher/gptlab-57m-3e17