2.0 KiB
2.0 KiB
license, base_model, pipeline_tag, tags
| license | base_model | pipeline_tag | tags | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| other | unsloth/Qwen2.5-14B-Instruct | text-generation |
|
Qwen2.5-14B-Instruct — iMatrix GGUF
GGUF imatrix quantizations of unsloth/Qwen2.5-14B-Instruct, published by Liodon AI.
Quick Start
llama.cpp
llama-cli -hf liodon-ai/Qwen2.5-14B-Instruct-imatrix-GGUF:Q4_K_M
Ollama
ollama run hf.co/liodon-ai/Qwen2.5-14B-Instruct-imatrix-GGUF:Q4_K_M
LM Studio / Jan — search liodon-ai/Qwen2.5-14B-Instruct-imatrix-GGUF and pick your quant.
Quants
| Quant | Size | VRAM est. | Notes |
|---|---|---|---|
IQ2_M |
5.36 GB | ~6 GB | 2-bit, iMatrix — smallest usable |
IQ3_M |
6.92 GB | ~8 GB | 3-bit, iMatrix — great quality/size tradeoff |
IQ4_XS |
8.12 GB | ~9 GB | 4-bit extra-small, iMatrix |
Q4_K_M |
8.99 GB | ~10 GB | 4-bit, iMatrix-calibrated (recommended) |
Q5_K_M |
10.51 GB | ~12 GB | 5-bit, iMatrix-calibrated |
Q6_K |
12.12 GB | ~14 GB | 6-bit, iMatrix-calibrated, near-lossless |
Q8_0 |
15.70 GB | ~18 GB | 8-bit, essentially lossless |
What is iMatrix?
Standard quantization treats all weights equally. iMatrix runs 128 calibration chunks through the full-precision model to find which weights matter most, then allocates more precision where it counts. At Q2/Q3/Q4 this means noticeably better coherence and instruction-following — same file size, better output.
Calibration: 2M tokens of WikiText-103.
Also see plain (non-iMatrix) quants:
liodon-ai/Qwen2.5-14B-Instruct-GGUF
Source
- Model: unsloth/Qwen2.5-14B-Instruct
- License: other
Quantized by Liodon AI