Instructions to use KoboldAI/fairseq-dense-13B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KoboldAI/fairseq-dense-13B with Transformers:

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

pipe = pipeline("text-generation", model="KoboldAI/fairseq-dense-13B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("KoboldAI/fairseq-dense-13B")
model = AutoModelForCausalLM.from_pretrained("KoboldAI/fairseq-dense-13B")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use KoboldAI/fairseq-dense-13B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KoboldAI/fairseq-dense-13B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KoboldAI/fairseq-dense-13B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/KoboldAI/fairseq-dense-13B

SGLang

How to use KoboldAI/fairseq-dense-13B 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 "KoboldAI/fairseq-dense-13B" \
    --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": "KoboldAI/fairseq-dense-13B",
		"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 "KoboldAI/fairseq-dense-13B" \
        --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": "KoboldAI/fairseq-dense-13B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use KoboldAI/fairseq-dense-13B with Docker Model Runner:
```
docker model run hf.co/KoboldAI/fairseq-dense-13B
```

This is a Hugging Face transformers-compatible conversion of the original dense 13B-parameter model from the paper "Efficient Large Scale Language Modeling with Mixtures of Experts" from Artetxe et al. Please refer to the original model card, which can be found at https://github.com/facebookresearch/fairseq/blob/main/examples/moe_lm/model_card.md.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	37.53
ARC (25-shot)	40.36
HellaSwag (10-shot)	75.51
MMLU (5-shot)	27.07
TruthfulQA (0-shot)	32.83
Winogrande (5-shot)	67.96
GSM8K (5-shot)	0.0
DROP (3-shot)	18.96

Downloads last month: 2,372

Spaces using KoboldAI/fairseq-dense-13B 29

Paper for KoboldAI/fairseq-dense-13B

Efficient Large Scale Language Modeling with Mixtures of Experts

Paper • 2112.10684 • Published Dec 20, 2021 • 2