unsloth
LFM2.5 8B A1B
LFM2.5 8B A1B is a 8.47B mixture-of-experts lfm2moe model from unsloth, with about 1.0B parameters active per token. Configured context length is 128,000 tokens. License: other. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters8.5B / 1.0B active
Context125K
Attentiongqa
Licenseother
24 layers · 8 KV heads · head dim 64 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 15.8 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 8.70 GB | Hugging Face file bytes |
| Q8_0 | 8.39 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 7.21 GB | Hugging Face file bytes |
| UD-Q6_K | 6.60 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 5.95 GB | Hugging Face file bytes |
| UD-Q5_K_M | 5.92 GB | Hugging Face file bytes |
| UD-Q5_K_S | 5.59 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 4.98 GB | Hugging Face file bytes |
| UD-Q4_K_M | 4.96 GB | Hugging Face file bytes |
| UD-Q4_K_S | 4.67 GB | Hugging Face file bytes |
| UD-IQ4_XS | 3.97 GB | Hugging Face file bytes |
| IQ4_NL | 4.06 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 3.73 GB | Hugging Face file bytes |
| Q3_K_M | 3.67 GB | Hugging Face file bytes |
| UD-IQ3_S | 3.33 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 3.05 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 2.72 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 2.52 GB | Hugging Face file bytes |
| UD-IQ1_M | 2.40 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Tight fit | UD-Q5_K_XL | 137–187 |
| GeForce RTX 3060 12GB | Tight fit | UD-Q8_K_XL | 132–179 |
| GeForce RTX 4060 Ti 16GB | Runs | UD-Q8_K_XL | 105–143 |
| GeForce RTX 4090 | Runs | BF16 | 231–314 |
| GeForce RTX 5090 | Runs great | BF16 | 410–559 |
| Apple M4 Pro | Tight fit | BF16 | 62–85 |
| Apple M5 | Tight fit | UD-Q8_K_XL | 56–76 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 20–28 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, UD-IQ4_XS, 102–139 tok/s
- GeForce RTX 2060 6GB — Tight fit, UD-IQ4_XS, 204–278 tok/s
- GeForce GTX 1660 Super — Tight fit, UD-IQ4_XS, 204–278 tok/s
- GeForce GTX 1660 Ti — Tight fit, UD-IQ4_XS, 175–238 tok/s
- Intel Arc A380 — Tight fit, UD-IQ4_XS, 113–154 tok/s
- GeForce RTX 5060 Ti 8GB — Tight fit, UD-Q5_K_XL, 213–290 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-Q8_K_XL).
llama.cpp
llama-server -hf unsloth/LFM2.5-8B-A1B-GGUF:UD-Q8_K_XL -c 8192 -ngl 24
Ollama
ollama run lfm2.5
LM Studio
lms get unsloth/LFM2.5-8B-A1B-GGUF # Then load quant UD-Q8_K_XL