---

base_model: reasonwang/SkillGym-Qwen3.5-9B

tags:

• gguf
• llama.cpp

---

SkillGym-Qwen3.5-9B-GGUF

A local GGUF build of reasonwang/SkillGym-Qwen3.5-9B for llama.cpp-compatible runtimes.

What you get

• SkillGym-Qwen3.5-9B-Q4_K_M.gguf
• Quantization: Q4_K_M
• Size: 5.24 GiB
• Ready for llama.cpp-compatible runtimes

The source checkpoint is multimodal, but this GGUF is a text-only conversion. Vision/image weights and an mmproj artifact are not included. Do not treat this file as preserving the source model's multimodal capability.

Run with llama.cpp

Command line

llama-cli -m SkillGym-Qwen3.5-9B-Q4_K_M.gguf -ngl 999 -c 8192

OpenAI-compatible server

llama-server -m SkillGym-Qwen3.5-9B-Q4_K_M.gguf -ngl 999 -c 8192

Then connect any OpenAI-compatible frontend to the llama.cpp server.

Choosing this quant

Q4_K_M is the published build in this repository. Larger quantizations preserve more of the original weights when available, while smaller ones reduce memory requirements.

Technical notes

Conversion

Conversion used --no-mtp because source config and source tensors disagreed about MTP/NextN presence.

Validation

Before upload this exact file passed a real llama-server /v1/chat/completions smoke gate. It checks model load, system/user separation, exact-output compliance, normal stop behavior, basic arithmetic, and chat-control-token leakage.

Reasoning mode used for smoke: off.

Source and integrity

Source revision used for conversion: 0b8f92556f6c7ec06bd858327ed2166da8b7697c.

Source HEAD checked immediately before publish: 0b8f92556f6c7ec06bd858327ed2166da8b7697c.

SHA256: b7e4d1f6adebf3499b60eb844954d7a601c83208ad31c7e330413cec1b84ae11

If source revisions differ while source safetensors fingerprints are unchanged, the difference is metadata-only and publishing remains allowed.

All model credit belongs to the original model authors.