Downloads · 30 days
15
13% of all-time downloads
faelfernandes/Qwen3.5-35B-A3B-3.5bit
Qwen3.5-35B-A3B-3.5bit is a machine learning model from faelfernandes. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
This is the Qwen3.5-35B-A3B model, natively quantized to 3.5bit for the MLX framework by Apple. This version allows running a massive model on standard Apple Silicon Macs without experiencing immediate OOM (Out Of Mem…
Downloads · 30 days
15
13% of all-time downloads
All-time downloads
113
Public
Parameters
34.7B
15.9 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors15.9 GB · 100%
How the weights are stored.
U3234.7B · 100%
From the Hugging Face model README
This is the Qwen3.5-35B-A3B model, natively quantized to 3.5bit for the MLX framework by Apple. This version allows running a massive model on standard Apple Silicon Macs without experiencing immediate OOM (Out Of Memory) issues.
O formato 3.5-bits (mixed_3_4) se revelou o Sweet Spot (Ponto de Ouro) para modelos de 35 Bilhões de parâmetros. Ele roda com tranquilidade num Mac de 24GB sem causar gargalos na GPU, ao mesmo tempo que retém a capacidade incrível de raciocínio lógico (Chain of Thought).
Durante os testes locais para validação dessa quantização, obtivemos os seguintes resultados de performance:
kIOGPUCommandBufferCallbackErrorTimeout (GPU Timeout) comum no Mac ao processar tensores gigantes, o processo de quantização deste modelo foi roteado cirurgicamente pela CPU. Isso garantiu integridade e sucesso total na conversão.oMLX is a high-performance serving engine for MLX models that exposes an OpenAI-compatible API.
1. Faça o download para o seu cache local:
python -c "from huggingface_hub import snapshot_download; snapshot_download('faelfernandes/Qwen3.5-35B-A3B-3.5bit')"
2. Inicie o servidor oMLX:
omlx serve --model-dir ~/.cache/huggingface/hub/models--faelfernandes--Qwen3.5-35B-A3B-3.5bit/snapshots/main
3. Teste via API (Padrão OpenAI):
curl http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen3.5-35B-A3B-3.5bit",
"messages": [{"role": "user", "content": "Explique a teoria da relatividade."}],
"temperature": 0.7
}'
1. Instale as dependências:
pip install mlx-lm huggingface_hub
2. Script de Inferência:
from mlx_lm import load, generate
model, tokenizer = load("faelfernandes/Qwen3.5-35B-A3B-3.5bit")
response = generate(model, tokenizer, prompt="Explain quantum mechanics in one paragraph.", max_tokens=500)
print(response)
Qwen/Qwen3.5-35B-A3BQuantizado e otimizado com foco na experiência do usuário de Mac. Por Rafael Fernandes 🇧🇷