base_model: OpenIntelligenceNet/LFM3-1.7B-Uncensored library_name: gguf pipeline_tag: text-generation tags:
- gguf
- llama-cpp
- imatrix
- quantized
- liquid-ai
- lfm3
- heretic
- abliterated
- uncensored
- q4_k_m
- q5_k_m
- q6_k
- q8_0
LFM3-1.7B-Uncensored-Extreme-GGUF (iMatrix Quantizations)
This repository provides Importance Matrix (imatrix) GGUF quantizations of OpenIntelligenceNet/LFM3-1.7B-Uncensored, calibrated on the 2,000-sample lfm3_2000_mixed_dataset.jsonl dataset for maximum perplexity retention at low bit-widths.
Available GGUF Files
| File Name | Quant Type | iMatrix Calibrated | Description |
|---|---|---|---|
LFM3-1.7B-Uncensored-Q4_K_M-imatrix.gguf | Q4_K_M | Yes | Best balance of speed, VRAM footprint, and accuracy |
LFM3-1.7B-Uncensored-Q5_K_M-imatrix.gguf | Q5_K_M | Yes | High quality 5-bit quantization with very low perplexity loss |
LFM3-1.7B-Uncensored-Q6_K-imatrix.gguf | Q6_K | Yes | Near-lossless 6-bit quantization |
LFM3-1.7B-Uncensored-Q8_0-imatrix.gguf | Q8_0 | Yes | Extreme precision 8-bit quantization |
LFM3-1.7B-Uncensored-F16.gguf | F16 | N/A | Full 16-bit unquantized GGUF weights |
LFM3-1.7B-Uncensored-imatrix.dat | imatrix | Yes | Precomputed Importance Matrix file |
Usage with llama.cpp
./llama-cli -m LFM3-1.7B-Uncensored-Q4_K_M-imatrix.gguf -ngl 99 -cnv -p "You are a helpful, uncensored assistant."