Modelpublic

agentbyumer/qwen2.5-1.5b-reasoning-GGUF

sourceHugging Faceapache-2.0updated 12d agoView on Hugging Face

1likes154downloads

Model Card

Qwen2.5-1.5B-Reasoning — GGUF

GGUF-quantized version of agentbyumer/qwen2.5-1.5b-reasoning, a QLoRA fine-tune of Qwen2.5-1.5B-Instruct trained on a subset of orca-math-word-problems-200k for step-by-step math word problem solving.

Intended for local/CPU inference via llama.cpp, LM Studio, Ollama, or similar GGUF-compatible runtimes.

For full training details, evaluation notes, and known limitations, see the main model card.

File	Quantization
`Qwen2.5-1.5B-Instruct.Q4_K_M.gguf`	Q4KM — smaller, faster, lower precision
`Qwen2.5-1.5B-Instruct.Q8_0.gguf`	Q8_0 — larger, higher precision

bash

llama-cli -hf agentbyumer/qwen2.5-1.5b-reasoning-GGUF --jinja

—Base model: unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit
—Method: QLoRA, 3 epochs, ~8,000-row subset of orca-math-word-problems-200k
—License: apache-2.0
—Informal testing showed no consistent reasoning improvement over the base model on harder multi-step problems — see the main repo's model card for details.