Requires a recent llama.cpp build. LFM2 is a hybrid (convolution + attention) architecture; only llama.cpp builds that include LFM2 support can load these files. Use a current release (or current LM Studio / Jan). Older builds will fail with an "unknown architecture 'lfm2'" error.

Files

File	Quant	Size	Notes
`LFM2.5-1.2B-Instruct-Uncensored-Q3_K_M.gguf`	Q3_K_M	573 MB	Smallest; lowest quality. For very tight memory.
`LFM2.5-1.2B-Instruct-Uncensored-Q4_K_M.gguf`	Q4_K_M	697 MB	Recommended — best size/quality balance.
`LFM2.5-1.2B-Instruct-Uncensored-Q5_K_M.gguf`	Q5_K_M	804 MB	Higher quality, slightly larger.
`LFM2.5-1.2B-Instruct-Uncensored-Q6_K.gguf`	Q6_K	918 MB	Near-lossless.
`LFM2.5-1.2B-Instruct-Uncensored-Q8_0.gguf`	Q8_0	1.2 GB	Effectively lossless vs the BF16 source.

Not sure which to pick? Start with Q4_K_M. Go up to Q5/Q6/Q8 if you have the memory and want maximum fidelity; drop to Q3 only if you're memory-constrained.

Usage

llama.cpp (auto-download from this repo)

bash

# Interactive chat — downloads the chosen quant automatically
llama-cli -hf zaakirio/LFM2.5-1.2B-Instruct-Uncensored-GGUF:Q4_K_M

# OpenAI-compatible server
llama-server -hf zaakirio/LFM2.5-1.2B-Instruct-Uncensored-GGUF:Q4_K_M -c 4096

Or, with a file you've already downloaded:

llama-cli -m LFM2.5-1.2B-Instruct-Uncensored-Q4_K_M.gguf -p "Hello, who are you?"

LM Studio / Jan

Search for zaakirio/LFM2.5-1.2B-Instruct-Uncensored-GGUF in the in-app model browser, or download a .gguf file from this page and load it.

Download a single file

bash

pip install -U "huggingface_hub[cli]"
hf download zaakirio/LFM2.5-1.2B-Instruct-Uncensored-GGUF \
  --include "LFM2.5-1.2B-Instruct-Uncensored-Q4_K_M.gguf" --local-dir ./

Prompt format

The chat template is embedded in the GGUF files, so chat-aware tools apply it automatically. For reference, it is ChatML-style:

<|startoftext|><|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

About the base model

This is a decensored derivative produced with Heretic (automatic directional ablation). Compared with the original LFM2.5-1.2B-Instruct:

Metric	Decensored	Original
Refusals (/100 harmful prompts)	5	98
KL divergence (harmless prompts)	0.1003	0 (by definition)

See the source model card for the full abliteration parameters and run details.

Intended use & disclaimer

This model has had its refusal behavior substantially removed and will comply with requests the original model would have declined. It is provided for research and unrestricted local use. You are responsible for how you use it and for complying with all applicable laws and with the base model's lfm1.0 license, which carries over to this derivative.

Provenance

Quantized from zaakirio/LFM2.5-1.2B-Instruct-Uncensored (BF16) using llama.cpp convert_hf_to_gguf.py + llama-quantize.
Base model: LiquidAI/LFM2.5-1.2B-Instruct
Decensoring tool: Heretic by p-e-w