Instructions to use DeepMount00/Mistral-RAG with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DeepMount00/Mistral-RAG with Transformers:

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

pipe = pipeline("text-generation", model="DeepMount00/Mistral-RAG")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DeepMount00/Mistral-RAG")
model = AutoModelForCausalLM.from_pretrained("DeepMount00/Mistral-RAG")

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use DeepMount00/Mistral-RAG with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DeepMount00/Mistral-RAG"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DeepMount00/Mistral-RAG",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/DeepMount00/Mistral-RAG

SGLang

How to use DeepMount00/Mistral-RAG 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 "DeepMount00/Mistral-RAG" \
    --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": "DeepMount00/Mistral-RAG",
		"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 "DeepMount00/Mistral-RAG" \
        --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": "DeepMount00/Mistral-RAG",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use DeepMount00/Mistral-RAG with Docker Model Runner:
```
docker model run hf.co/DeepMount00/Mistral-RAG
```

Mistral-RAG / README.md

DeepMount00

Update README.md

75ce554 verified about 2 years ago

preview code

raw

history blame contribute delete

3.69 kB

metadata

library_name: transformers
license: apache-2.0
datasets:
  - DeepMount00/gquad_it
language:
  - it

Mistral-RAG

Model Name: Mistral-RAG
Base Model: Mistral-Ita-7b
Specialization: Question and Answer Tasks

Overview

Mistral-RAG is a refined fine-tuning of the Mistral-Ita-7b model, engineered specifically to enhance question and answer tasks. It features a unique dual-response capability, offering both generative and extractive modes to cater to a wide range of informational needs.

Capabilities

Generative Mode

Description: The generative mode is designed for scenarios that require complex, synthesized responses. This mode integrates information from multiple sources and provides expanded explanations.
Ideal Use Cases:
- Educational purposes
- Advisory services
- Creative scenarios where depth and detailed understanding are crucial

Extractive Mode

Description: The extractive mode focuses on speed and precision. It delivers direct and concise answers by extracting specific data from texts.
Ideal Use Cases:
- Factual queries in research
- Legal contexts
- Professional environments where accuracy and direct evidence are necessary

How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

MODEL_NAME = "DeepMount00/Mistral-RAG"

model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.bfloat16).eval()
model.to(device)
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

def generate_answer(prompt, response_type="generativo"):
    # Creazione del contesto e della domanda in base al tipo di risposta
    if response_type == "estrattivo":
        prompt = f"Rispondi alla seguente domanda in modo estrattivo, basandoti esclusivamente sul contesto.\n{prompt}"
    else:
        prompt = f"Rispondi alla seguente domanda in modo generativo, basandoti esclusivamente sul contesto.\n{prompt}"

    # Preparazione del messaggio per il modello
    messages = [
        {"role": "user", "content": prompt},
    ]
    model_inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(device)
    generated_ids = model.generate(model_inputs, max_new_tokens=200, do_sample=True,
                                   temperature=0.001, eos_token_id=tokenizer.eos_token_id)
    decoded = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    return decoded[0].split("[/INST]", 1)[1].strip() if "[/INST]" in decoded[0] else "Errore nella generazione della risposta"



# Esempio di utilizzo con la nuova funzionalità
contesto = """Venerdì più di 2.100 persone che vivono vicino a un vulcano in Indonesia sono state sfollate per i rischi legati a un’eruzione. Martedì infatti l’isola vulcanica di Ruang, che si trova circa 100 chilometri a nord di Sulawesi, ha cominciato a eruttare, producendo una colonna di fumo e ceneri che ieri ha raggiunto 1.200 metri di altezza. Le operazioni di evacuazione sono ancora in corso: complessivamente sono più di 11mila le persone a cui è stato detto di lasciare le proprie case. Gran parte di loro vive sulla vicina isola di Tagulandang, che in totale ha 20mila abitanti; potrebbe essere raggiunta non solo dalle ceneri vulcaniche e dai piroclasti, ma anche da un eventuale tsunami causato dalla caduta in mare di lava e rocce."""
domanda = "Perchè le persone sono evacuate dalle case?"
prompt = f"Contesto: {contesto}\nDomanda: {domanda}"


answer = generate_answer(prompt, "estrattivo")
print(answer)

Developer

[Michele Montebovi]