DJLougen/MiroThinker-1.7-mini-GGUF-Q4_K_M

MiroThinker-1.7-mini GGUF Q4_K_M

Q4_K_M GGUF quantization of miromind-ai/MiroThinker-1.7-mini.

Model Details

• Original Model: miromind-ai/MiroThinker-1.7-mini (Qwen3 MoE, 30.5B params)
• Quantization: Q4_K_M (4-bit mixed)
• File Size: ~18 GB
• Format: GGUF (llama.cpp compatible)
• Max Context: 256K tokens
• Max Tool Calls: 300

About MiroThinker-1.7-mini

MiroThinker-1.7-mini is a deep research agent model fine-tuned from Qwen3-30B-A3B-Thinking-2507. It achieves state-of-the-art performance in deep research tasks among open-source models.

Benchmarks (original BF16)

Benchmark	Score
BrowseComp	74.0%
BrowseComp-ZH	75.3% (SOTA)
GAIA-Val-165	82.7%
HLE-Text	42.9%

Usage

Works with any GGUF-compatible runtime: llama.cpp, Ollama, LM Studio, etc.

Ollama:

ollama run hf.co/DJLougen/MiroThinker-1.7-mini-GGUF-Q4_K_M

llama.cpp:

llama-cli -m MiroThinker-1.7-mini-Q4_K_M.gguf -c 8192 -n 512

Recommended Parameters

• temperature: 1.0
• top_p: 0.95
• repetition_penalty: 1.05

Credits

• Original model by miromind-ai
• Quantized by DJLougen using llama.cpp