LFM2.5 8B A1B heretic needs about 7.3 GB of memory at Q4_K_M (a 5.2 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 5 quants, from 5.2 GB (Q4_K_M) to 16.9 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 6.4 GB.
LFM2.5 8B A1B heretic – a decensored, on‑device MoE model with 128K context
LFM2.5 8B A1B heretic is a 8.47B‑parameter hybrid conv+attention MoE model. It supports reasoning, roleplay and tool use, and offers a 128,000‑token context window. The decensored variant removes refusal behaviour while keeping the base model’s agentic abilities.
Good at
- Reasoning with chain of thought
- Pythonic tool‑call format
- 128K context length
Limits
- No vision support
- License is other
- Requires fast RAM for MoE bandwidth
Best for
- On‑device personal assistants
- Agentic workflows with tool use
- Multilingual text generation
Who it’s for. Developers who need a fast, uncensored on‑device assistant that can call tools. It runs at interactive speed on CPU‑only laptops, Apple Silicon, and low‑VRAM GPUs while fitting a 128K context.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 5.2 GB | 7.3 GB | Tight | Get file → | |
| Q5_K_M | 6.0 GB | 8.1 GB | Too big | Get file → | |
| Q6_K | 7.0 GB | 9.1 GB | Too big | Get file → | |
| Q8_0 | 9.0 GB | 11.1 GB | Too big | Get file → | |
| F16 | 16.9 GB | 19.1 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_M (5.16 GB file, ~7.3 GB VRAM at 8 k context). If more memory is available, Q5_K_M (6.03 GB, ~8.1 GB VRAM) offers better quality. Higher quants such as Q6_K, Q8_0 or F16 increase VRAM needs to 9.1 GB, 11.1 GB or 19.1 GB respectively.
What GPU do you need for LFM2.5 8B A1B heretic?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: the best quant that fits is Q4_K_M (5.2 GB file, about 7.3 GB in use).
- 12 GB VRAM: the best quant that fits is Q8_0 (9.0 GB file, about 11.1 GB in use).
- 16 GB VRAM: the best quant that fits is Q8_0 (9.0 GB file, about 11.1 GB in use).
- 24 GB VRAM: the best quant that fits is F16 (16.9 GB file, about 19.1 GB in use).
Which LFM2.5 8B A1B heretic quant should you choose?
Q4_K_M is the usual starting point for LFM2.5 8B A1B heretic: it balances file size and quality well. Here it is a 5.2 GB file.
The Q8_0 file is 9.0 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about LFM2.5 8B A1B heretic
How much VRAM does LFM2.5 8B A1B heretic need?
About 7.3 GB at Q4_K_M with an 8K context window. That is the 5.2 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run LFM2.5 8B A1B heretic on an 8 GB GPU?
Yes, with Q4_K_M: the 5.2 GB file needs about 7.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of LFM2.5 8B A1B heretic GGUF?
It depends on the quant: Q4_K_M is 5.2 GB, Q5_K_M is 6.0 GB, Q6_K is 7.0 GB, Q8_0 is 9.0 GB, F16 is 16.9 GB.
What context length does LFM2.5 8B A1B heretic support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 1.6 GB at 8K and 6.4 GB at 32K.
Where can I download LFM2.5 8B A1B heretic GGUF?
The files are on Hugging Face in the saidutta69/LFM2.5-8B-A1B-heretic repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does LFM2.5 8B A1B heretic use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Does this model support image input?
No, it is a text‑only model with no vision capability.
What is the context length of this model?
The model was trained with a 128,000‑token context length.