Instructions to use stage-babylm/llama-256-8L with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use stage-babylm/llama-256-8L with Transformers:

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

pipe = pipeline("text-generation", model="stage-babylm/llama-256-8L")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("stage-babylm/llama-256-8L")
model = AutoModelForCausalLM.from_pretrained("stage-babylm/llama-256-8L", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use stage-babylm/llama-256-8L with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "stage-babylm/llama-256-8L"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "stage-babylm/llama-256-8L",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/stage-babylm/llama-256-8L

SGLang

How to use stage-babylm/llama-256-8L 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 "stage-babylm/llama-256-8L" \
    --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": "stage-babylm/llama-256-8L",
		"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 "stage-babylm/llama-256-8L" \
        --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": "stage-babylm/llama-256-8L",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use stage-babylm/llama-256-8L with Docker Model Runner:
```
docker model run hf.co/stage-babylm/llama-256-8L
```

llama-256-8L

This model is a fine-tuned version of on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.7280

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0018
train_batch_size: 32
eval_batch_size: 32
seed: 42
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.95) and epsilon=1e-06 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 0.05
num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0	0	6.9575
2.3785	0.0500	2014	2.3882
2.1737	0.1000	4028	2.1826
2.0899	0.1500	6042	2.1034
2.0427	0.2000	8056	2.0623
2.0186	0.2500	10070	2.0367
1.9918	0.3000	12084	2.0073
1.9593	0.3500	14098	1.9862
1.9399	0.4000	16112	1.9662
1.9321	0.4500	18126	1.9422
1.8977	0.5000	20140	1.9194
1.8826	0.5500	22154	1.8998
1.8431	0.6000	24168	1.8738
1.8327	0.6500	26182	1.8467
1.8078	0.7000	28196	1.8225
1.7715	0.7500	30210	1.7944
1.7468	0.8000	32224	1.7725
1.7327	0.8500	34238	1.7515
1.7122	0.9000	36252	1.7370
1.7045	0.9500	38266	1.7296
1.7016	1.0	40278	1.7280

Framework versions

Transformers 5.14.1
Pytorch 2.13.0+cu130
Datasets 5.0.0
Tokenizers 0.22.2

Downloads last month: 1,592

Safetensors

Model size

6.8M params

Tensor type

F32