Instructions to use djelia/bm-mistral-7b-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use djelia/bm-mistral-7b-base with Transformers:

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

pipe = pipeline("text-generation", model="djelia/bm-mistral-7b-base")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("djelia/bm-mistral-7b-base")
model = AutoModelForCausalLM.from_pretrained("djelia/bm-mistral-7b-base", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use djelia/bm-mistral-7b-base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "djelia/bm-mistral-7b-base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "djelia/bm-mistral-7b-base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/djelia/bm-mistral-7b-base

SGLang

How to use djelia/bm-mistral-7b-base 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 "djelia/bm-mistral-7b-base" \
    --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": "djelia/bm-mistral-7b-base",
		"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 "djelia/bm-mistral-7b-base" \
        --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": "djelia/bm-mistral-7b-base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Unsloth Studio

How to use djelia/bm-mistral-7b-base with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for djelia/bm-mistral-7b-base to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for djelia/bm-mistral-7b-base to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for djelia/bm-mistral-7b-base to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="djelia/bm-mistral-7b-base",
    max_seq_length=2048,
)

Docker Model Runner
How to use djelia/bm-mistral-7b-base with Docker Model Runner:
```
docker model run hf.co/djelia/bm-mistral-7b-base
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

bm-mistral-7b-base

Mistral 7B v0.3 adapted to Bambara (Bamanankan). A LoRA run with r=128 over all seven attention and MLP projections, with embed_tokens and lm_head retrained in full — the shape of vocabulary and language adaptation rather than a task fine-tune.

MistralForCausalLM, bfloat16 — 32 layers, hidden size 4096, intermediate size 14336, 32 attention heads with 8 KV heads (GQA), head dim 128, 32,768-token vocabulary, 32,768-token context, RoPE theta 1e6, no sliding window. Tokenizer is LlamaTokenizer with padding_side="left".

Usage

The repo ships merged weights alongside the LoRA adapter; load the merged weights with AutoModelForCausalLM.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "djelia/bm-mistral-7b-base"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)

prompt = "Bamanankan ye kan ye min bɛ fɔ"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(
    **inputs,
    max_new_tokens=128,
    do_sample=True,
    temperature=0.7,
    top_p=0.9,
    pad_token_id=tokenizer.pad_token_id,
)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Prompt format

This is a base checkpoint for raw Bambara continuation and for further fine-tuning — there is no chat template, so apply_chat_template does not apply. The instruction-tuned sibling djelia/bm-mistral-7b-v1 adds the Bambara Alpaca layout (### ɲɛfɔli: / ### Donnafɛnw: / ### Jaabi:) on top of the same architecture.

generation_config.json carries only the token ids and max_length, so pass your own sampling settings.

Downloads last month: 1

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for djelia/bm-mistral-7b-base

Base model

mistralai/Mistral-7B-v0.3

Quantized

unsloth/mistral-7b-v0.3-bnb-4bit

Finetuned

(643)

this model