Instructions to use AgentPublic/fabrique-reference-2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AgentPublic/fabrique-reference-2 with Transformers:

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

pipe = pipeline("text-generation", model="AgentPublic/fabrique-reference-2")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AgentPublic/fabrique-reference-2")
model = AutoModelForCausalLM.from_pretrained("AgentPublic/fabrique-reference-2")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use AgentPublic/fabrique-reference-2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AgentPublic/fabrique-reference-2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AgentPublic/fabrique-reference-2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AgentPublic/fabrique-reference-2

SGLang

How to use AgentPublic/fabrique-reference-2 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 "AgentPublic/fabrique-reference-2" \
    --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": "AgentPublic/fabrique-reference-2",
		"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 "AgentPublic/fabrique-reference-2" \
        --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": "AgentPublic/fabrique-reference-2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use AgentPublic/fabrique-reference-2 with Docker Model Runner:
```
docker model run hf.co/AgentPublic/fabrique-reference-2
```

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

The fabrique-reference-2 model features three different modes:

Simple mode: this corresponds to the original etalab model (fabrique-miaou). The answer is created from an experience, and ideally from answer elements and links.
Experience mode: the model relies on a RAG on the experiences of servicespublics+, searching for similar experiences (from the same institution) and integrating the answers into the prompt. With this approach, you don't need to provide anything other than the answer and the institution. Provided you already have similar responses, this mode should solve most hallucinations linked to contact data and links.
Expert mode: the model relies on a RAG on the expert documents servicespublics.fr or, optionally, other reference sources. It is necessary to have already generated an initial response, as this mode does not handle other aspects specific to servicespublics+ (user interaction, contact data, etc.). The final text will be enriched with advanced information and standardized references based on the Wikipedia model (with tags).

In terms of design, the user's journey can go through two stages:

Simple mode or Experience mode to create the first response to the experience in the style of servicespublics+.
Expert mode to add referenced information to the response.

Prompt

Downloads last month: 8

Safetensors

Model size

13B params

Tensor type

BF16