No official torrent yet — requesting it tells the flock what to host next. You cannot seed a catalog-only page.
Cataloged from froggeric/Qwen3.6-27B-MLX-4bit
All models by froggeric →
Qwen3.6-27B
MLX 4-bit · Text + Vision + Thinking + Tool Calling
Apple Silicon native
Qwen3.6-27B is a 27B-parameter dense model from Alibaba. It uses a hybrid linear/full attention architecture (3:1 ratio across 64 layers) that combines efficient DeltaNet-style linear attention with full softmax attention at regular intervals. It supports 262K context, vision, video, and multi-token prediction.
This is an MLX 4-bit conversion of the official Qwen3.6-27B weights, ready to run on Apple Silicon with full text, image, and video support. At 4.6 bits/weight, this fits in 16 GB of memory — ideal for 24 GB Macs.
mlx-vlm. Text, image, and video inputs work out of the box<|think_on|> / <|think_off|> tags (see below)mlx-lm >= 0.31.2, mlx-vlm >= 0.4.4The first line of your system prompt must be:
The model underperforms without it. You can append anything after that line.
This model ships with a fixed Jinja chat template that lets you toggle thinking on the fly. Drop <|think_on|> or <|think_off|> anywhere in your system or user prompt. The template intercepts the tag, strips it from context so the model never sees it, and flips the thinking mode.
Fast answer, no internal reasoning.
The model thinks step by step, then answers.
The bundled Jinja template fixes several issues in the official Qwen 3.6 template:
|items filter and |safe, which do not exist in C++ Jinja runtimes (LM Studio, MLX). This template uses direct dictionary key lookups instead.developer role crashes. Modern APIs send message.role == "developer". The official template throws an exception. This template maps it to system.preserve_thinking spam. The official template wraps every past turn in empty <think/> blocks, wasting context tokens. This template only emits thinking blocks when they contain actual reasoning content.</thinking> hallucination handling. The model sometimes generates </thinking> instead of the expected closing tag. This template handles both gracefully.<|think_on|> / <|think_off|> from any message role.See chat_template.README.md [blocked] for the full breakdown.
From the official Qwen authors. Reserve 128K+ context for thinking mode.
GGUF runtimes use presence_penalty (0 = off). MLX / LM Studio use repeat_penalty (1.0 = off).
Apache-2.0, inherited from Qwen3.6.
from mlx_lm import load, generate
model, tokenizer = load("froggeric/Qwen3.6-27B-MLX-4bit")response = generate(model, tokenizer, prompt="Hello", max_tokens=256, temp=0.7)print(response)from mlx_vlm import load, generatefrom mlx_vlm.prompt_utils import apply_chat_template
model, processor = load("froggeric/Qwen3.6-27B-MLX-4bit")image = ["path/to/image.jpg"]prompt = "Describe this image."formatted = apply_chat_template(processor, model.config, prompt, num_images=len(image))result = generate(model, processor, formatted, image, max_tokens=256, temp=0.7)print(result.text)# Textmlx_lm.generate --model froggeric/Qwen3.6-27B-MLX-4bit --prompt "Hello"
# Visionmlx_vlm.generate --model froggeric/Qwen3.6-27B-MLX-4bit --image image.jpg --prompt "Describe this image"You are Qwen, created by Alibaba Cloud. You are a helpful assistant.System: You are a coding assistant. <|think_off|>User: What's 2+2?System: You are a coding assistant. <|think_on|>User: Implement a red-black tree in Rust.