LFM2.5 350M needs about 1.3 GB of memory at Q4_K_M (a 0.2 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 12 quants, from 0.2 GB (Q2_K) to 0.7 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.1 GB.
LFM2.5 350M – a 350M parameter hybrid model for on‑device text generation
LFM2.5 350M is a 350M parameter text‑only model built on the LFM2 architecture. It supports tool use and runs under 1 GB of memory with llama.cpp, MLX, and vLLM. The model was trained on 28T tokens and offers a 32,768 token context length.
Good at
- Tool‑enabled text generation
- Low‑memory inference
- Fast edge decoding
Limits
- No vision support
- Not suited for knowledge‑intensive tasks
- No programming assistance
Best for
- Data extraction
- Structured output creation
- Function calling
Who it’s for. Developers needing a lightweight, on‑device LLM that can perform tool calls and structured generation. Ideal for mobile or embedded applications where memory is limited.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 0.2 GB | 1.2 GB | Fits | Get file → | |
| Q3_K_S | 0.2 GB | 1.2 GB | Fits | Get file → | |
| Q3_K_M | 0.2 GB | 1.2 GB | Fits | Get file → | |
| Q3_K_L | 0.2 GB | 1.2 GB | Fits | Get file → | |
| IQ4_XS | 0.2 GB | 1.2 GB | Fits | Get file → | |
| Q4_K_S | 0.2 GB | 1.3 GB | Fits | Get file → | |
| Q4_K_M | 0.2 GB | 1.3 GB | Fits | Get file → | |
| Q5_K_S | 0.3 GB | 1.3 GB | Fits | Get file → | |
| Q5_K_M | 0.3 GB | 1.3 GB | Fits | Get file → | |
| Q6_K | 0.3 GB | 1.3 GB | Fits | Get file → | |
| Q8_0 | 0.4 GB | 1.4 GB | Fits | Get file → | |
| F16 | 0.7 GB | 1.7 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_M (0.23 GB file, ~1.3 GB memory) or Q5_K_M (0.26 GB file, ~1.3 GB memory). For 4 GB VRAM, Q3_K_M (0.19 GB file, ~1.2 GB memory) or Q3_K_S (0.18 GB file, ~1.2 GB memory) are suitable. Higher precision quants like F16 (0.71 GB file, ~1.7 GB memory) require more VRAM.
What GPU do you need for LFM2.5 350M?
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: the best quant that fits is F16 (0.7 GB file, about 1.7 GB in use).
Which LFM2.5 350M quant should you choose?
Q4_K_M is the usual starting point for LFM2.5 350M: it balances file size and quality well. Here it is a 0.2 GB file.
The Q8_0 file is 0.4 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.
If memory is tight, the smallest file is Q2_K at 0.2 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about LFM2.5 350M
How much VRAM does LFM2.5 350M need?
About 1.3 GB at Q4_K_M with an 8K context window. That is the 0.2 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 350M on an 8 GB GPU?
Yes, with F16: the 0.7 GB file needs about 1.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of LFM2.5 350M GGUF?
It depends on the quant: Q2_K is 0.2 GB, Q3_K_S is 0.2 GB, Q3_K_M is 0.2 GB, Q3_K_L is 0.2 GB, IQ4_XS is 0.2 GB, Q4_K_S is 0.2 GB and 6 more.
What context length does LFM2.5 350M support?
The GGUF header lists a trained context length of 128,000 tokens (125K). The KV cache grows with the context you use: about 0.5 GB at 8K and 2.1 GB at 32K.
Where can I download LFM2.5 350M GGUF?
The files are on Hugging Face in the mradermacher/LFM2.5-350M-GGUF 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 350M 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.
Can I run LFM2.5 350M on 8 GB VRAM?
Yes, quants such as Q4_K_M or Q5_K_M use about 1.3 GB of VRAM at 8 k context, well within an 8 GB budget.
Does this model support image input?
No, LFM2.5 350M is a text‑only model and does not support vision.
What is the maximum context length?
The model supports up to 32,768 tokens of context.