BigBang-v1-MLX-Q4_K_M

This model was converted to MLX format from endless-frontier/BigBang-v1 using mlx_lm version 0.31.3.

Quantization: mixed_4_6 (equivalent to llama.cpp Q4_K_M). 4-bit base with higher bits applied to first/last 1/8 of layers, every 3rd layer, v_proj/down_proj, and lm_head.

Use with mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/BigBang-v1-MLX-Q4_K_M")

messages = [{"role": "user", "content": "Hello!"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
response = generate(model, tokenizer, prompt=prompt, max_tokens=512)

Or via CLI:

python3 -m mlx_lm.chat --model mlx-community/BigBang-v1-MLX-Q4_K_M

Model Info

• Architecture: qwen3_5_moe — 35B total, 3B active, 256 experts (8/tok)
• Attention: Hybrid linear + full attention
• Context: 262K tokens
• Type: Text-only (language model)

Variants

License

Apache 2.0 — Copyright 2026 Alibaba Cloud (original base model).