Qwen2.5-Coder-7B-Instruct — GGUF Quants

Quantized GGUF versions of Qwen/Qwen2.5-Coder-7B-Instruct — Alibaba's specialized code generation model trained on 5.5 trillion tokens of code data. Achieves performance comparable to much larger general models on coding benchmarks — the go-to for local code assistance.

Available Files

File	Quant	Size	Use Case
`Qwen2.5-Coder-7B-Instruct-Q8_0.gguf`	Q8_0	~7.7GB	Maximum quality
`Qwen2.5-Coder-7B-Instruct-Q6_K.gguf`	Q6_K	~6.0GB	Near-lossless
`Qwen2.5-Coder-7B-Instruct-Q5_K_M.gguf`	Q5_K_M	~5.2GB	High quality
`Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf`	Q4_K_M	~4.4GB	Recommended default
`Qwen2.5-Coder-7B-Instruct-Q3_K_M.gguf`	Q3_K_M	~3.5GB	Low VRAM
`Qwen2.5-Coder-7B-Instruct-IQ4_XS.gguf`	IQ4_XS	~3.9GB	Imatrix 4-bit
`Qwen2.5-Coder-7B-Instruct-IQ3_XXS.gguf`	IQ3_XXS	~2.9GB	Imatrix 3-bit
`Qwen2.5-Coder-7B-Instruct-IQ2_M.gguf`	IQ2_M	~2.5GB	Imatrix 2-bit
`Qwen2.5-Coder-7B-Instruct-IQ1_S.gguf`	IQ1_S	~1.8GB	Extreme compression
`Qwen2.5-Coder-7B-Instruct-fp16.gguf`	FP16	~14.8GB	Full precision
`imatrix.dat`	—	—	Importance matrix

Usage

./llama-cli -m Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf \
  --ctx-size 8192 -n 1024 \
  -p "<|im_start|>system\nYou are an expert programmer.<|im_end|>\n<|im_start|>user\nWrite a Python function to sort a list.<|im_end|>\n<|im_start|>assistant\n"

About Qwen2.5-Coder-7B

Parameters: 7B
Training: 5.5T tokens of code data
Context: 128K tokens
License: Apache 2.0
Strengths: Code completion, debugging, code explanation, FIM for IDE integrations

Best-in-class local code assistance at the 7B scale.

Quantized by DuoNeural using llama.cpp on RTX 5090.

DuoNeural

DuoNeural is an open AI research lab — human + AI in collaboration.

Platform	Link
HuggingFace	huggingface.co/DuoNeural
Website	duoneural.com
GitHub	github.com/DuoNeural
X / Twitter	@DuoNeural
Email	duoneural@proton.me
Newsletter	duoneural.beehiiv.com
Support	buymeacoffee.com/duoneural

DuoNeural Research Publications

Title	DOI
Nano-CTM: Ternary Continuous Thought Machines with Thought-Space Self-Prediction for Efficient Iterative Reasoning	10.5281/zenodo.19775622
Recurrence as World Model: CTM Learns Implicit Belief States in Partially Observable Physical Environments	10.5281/zenodo.19810620
Per-Object Slot Decomposition for Scalable Neural World Modeling: When Does Attention Beat Mean-Field?	10.5281/zenodo.19846804
The Dynamical Horizon Principle: CTM Gates Converge to the Predictability Limit of Dynamical Systems	10.5281/zenodo.19952612