Qwen3.6-30B-RYS-Bombaklot-Q4_K_M

Update: I noticed a training bug in my original upload. I corrected it and re-uploaded the corrected weights. Please re-download if you downloaded this model previously.

Qwen3.6-30B-RYS-Bombaklot (GGUF Q4_K_M)

This repository contains the Q4_K_M GGUF quantized format of the Qwen3.6-30B-RYS-Bombaklot model.

This model has been specifically designed and optimized for agentic coding, agentic harnesses, advanced mathematics, and logic-heavy tasks, leveraging enhanced bi-directional logical reasoning capabilities within the latent space.

New Capabilities

By deepening the model architecture and training the model with bi-directional logical reasoning, the model unlocks advanced cognitive and logical processing flows. It evaluates problems from multiple angles and revisits context iteratively, making it exceptionally powerful for complex problem-solving, multi-step logical deductions, and operating within autonomous agent frameworks.

Because this is a reasoning-heavy model, set your thinking budget to at least 8,192 tokens, and 16,384 tokens if working on a very complex agentic, math, or coding workflow.

Files Available

Preferred Sampling Parameters

For precise coding and logic tasks, we highly recommend the following base sampling parameters:

• Temperature: 0.6
• Top-P: 0.95
• Top-K: 20
• Repetition Penalty: 1.0

Usage with llama.cpp

To run the model, you will need a relatively recent build of llama.cpp that supports the Qwen3.6/3.5 architectures.

CLI Usage

./llama-cli -m Qwen3.6-30B-RYS-Bombaklot.Q4_K_M.gguf \
  --color \
  -c 32768 \
  --temp 0.6 \
  --top-p 0.95 \
  --top-k 20 \
  --repeat-penalty 1.0 \
  -p "<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n<|im_start|>user\nHello! Can you write a Python script for me?<|im_end|>\n<|im_start|>assistant\n<think>\n"

Server Usage

You can also run it via the local API server for high-throughput interactions:

./llama-server -m Qwen3.6-30B-RYS-Bombaklot.Q4_K_M.gguf \
  --port 8080 \
  -c 32768 \
  --temp 0.6 \
  --top-p 0.95 \
  --top-k 20

Citation & Acknowledgements

The base model utilizes the RYS (Repeat YourSelf) technique, credited to its creator dnhkng. We also extend our gratitude to the Qwen team at Alibaba Cloud for their foundational Qwen series open weights.