LFM 2.5 24B A2B GGUF Size & VRAM Requirements
LFM2.5-24B-A2B-GGUF is a large language model focused on general purpose & instruction. With approximately 23.97 billion parameters, it balances local hardware requirements with reasoning and generation fidelity. It is built on the lfm2moe architecture with a native context window of up to 131,072 tokens. The model is primarily recommended for general knowledge, document summarization, email drafting, language translation, and multi-turn conversational chat. Released under the other license, it can be executed fully offline without sending data to external APIs.
LFM2.5-24B-A2B-GGUF is a Mixture-of-Experts model with 64 experts, of which 4 are active on each token. Routing only 4 of 64 experts makes it noticeably faster than a dense model of the same size — but every expert still has to be held in memory, so the numbers below are set by the full parameter count, not the active one. It is built from 40 transformer layers, a hidden size of 2,048, 32 attention heads. It uses a hybrid attention design: only 10 of its 40 layers keep a full KV cache, while the rest use linear attention with a constant-size state. Long contexts therefore cost far less memory than on a conventional model of this size.
To run unsloth/LFM2.5-24B-A2B-GGUF locally at a 4,096-token context, its quantized versions need between 7.83 GB (IQ2_XXS, lowest quality) and 45.55 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for LFM2.5-24B-A2B-GGUF is about 0.08 GB, rising to roughly 2.5 GB at its full 131,072-token context. Shorter prompts free up memory for a higher-quality quantization.
For most users the best balance is IQ2_M, needing about 7.9 GB. That means unsloth/LFM2.5-24B-A2B-GGUF fits entirely in the VRAM of an 8 GB GPU or larger, running fully on the GPU.
Available GGUF quantizations for unsloth/LFM2.5-24B-A2B-GGUF include IQ2_XXS, IQ2_M, Q2_K_XL, IQ3_XXS, Q3_K_XL, Q4_K_XL, Q5_K_XL, Q6_K_XL, Q8_0, BF16. The model supports a native context length of up to 131,072 tokens; a longer context grows the KV cache and the memory needed.
Recommended Use Cases & Local Setup
Ideal for: General knowledge, document summarization, email drafting, language translation, and multi-turn conversational chat.
Simple one-line CLI installation running silently in the background
Polished desktop client with one-click model downloads and GPU offloading
Open-source privacy-focused desktop assistant
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| IQ2_XXS | 2.49 | Very low | 6.96 GB | 0.08 GB | 7.83 GB | 57.5 t/s | Fits in VRAM |
| IQ2_M | 2.51 | Very low | 7.02 GB | 0.08 GB | 7.9 GB | 57.0 t/s | Fits in VRAM |
| Q2_K_XL | 2.68 | Low | 7.49 GB | 0.08 GB | 8.37 GB | 6.7 t/s | Offload |
| IQ3_XXS | 2.99 | Low | 8.34 GB | 0.08 GB | 9.22 GB | 6.0 t/s | Offload |
| Q3_K_XL | 3.65 | Fair | 10.2 GB | 0.08 GB | 11.07 GB | 4.9 t/s | Offload |
| Q4_K_XL | 4.96 | Good | 13.86 GB | 0.08 GB | 14.73 GB | 3.6 t/s | Offload |
| Q5_K_XL | 5.98 | Very good | 16.69 GB | 0.08 GB | 17.57 GB | 3.0 t/s | Offload |
| Q6_K_XL | 7.3 | Excellent | 20.36 GB | 0.08 GB | 21.24 GB | 2.5 t/s | Offload |
| Q8_0 | 8.51 | Excellent | 23.74 GB | 0.08 GB | 24.62 GB | — | Insufficient |
| BF16 | 16.01 | Excellent | 44.67 GB | 0.08 GB | 45.55 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is unsloth/LFM2.5-24B-A2B-GGUF?
unsloth/LFM2.5-24B-A2B-GGUF is a language model with 23.97 billion parameters, based on the lfm2moe architecture. It is released under the other license and distributed as GGUF files for local inference.
Is unsloth/LFM2.5-24B-A2B-GGUF a Mixture-of-Experts (MoE) model?
Yes. unsloth/LFM2.5-24B-A2B-GGUF is a Mixture-of-Experts model with 64 experts, of which 4 are activated per token. That makes it faster than a dense model of the same size, but all 64 experts must be loaded into memory, so the VRAM/RAM it needs is driven by the total parameter count, not the active one.
How does unsloth/LFM2.5-24B-A2B-GGUF handle long context?
unsloth/LFM2.5-24B-A2B-GGUF uses a hybrid attention design: only 10 of its 40 layers keep a full KV cache, and the rest use linear attention with a constant-size state. Long contexts cost far less memory than on a conventional model of this size.
How much VRAM do you need to run unsloth/LFM2.5-24B-A2B-GGUF?
You need about 7.9 GB of VRAM to run unsloth/LFM2.5-24B-A2B-GGUF entirely on the GPU using the IQ2_M quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run unsloth/LFM2.5-24B-A2B-GGUF on an 8 GB GPU?
Yes. With 8 GB of VRAM you can run unsloth/LFM2.5-24B-A2B-GGUF fully on the GPU using IQ2_M (about 7.9 GB).
Can I run unsloth/LFM2.5-24B-A2B-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run unsloth/LFM2.5-24B-A2B-GGUF fully on the GPU using Q4_K_XL (about 14.73 GB).
Can I run unsloth/LFM2.5-24B-A2B-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run unsloth/LFM2.5-24B-A2B-GGUF fully on the GPU using Q6_K_XL (about 21.24 GB).
What context length does unsloth/LFM2.5-24B-A2B-GGUF support?
unsloth/LFM2.5-24B-A2B-GGUF supports a native context length of up to 131,072 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for unsloth/LFM2.5-24B-A2B-GGUF?
For unsloth/LFM2.5-24B-A2B-GGUF, higher bits-per-weight means better quality. A common sweet spot is a Q4_K_M or Q5_K_M quantization, which keeps most of the quality while roughly halving the memory versus 8-bit. Pick the highest quantization that still fits in your VRAM.