LFM2.5-8B-A1B
Files and memory at 8K context
| Quant | File | Files | + KV + runtime | Smallest fit | Source |
|---|---|---|---|---|---|
| Q4_K_M LiquidAI/LFM2.5-8B-A1B-GGUF | 4.80 GiB | 1 | 5.9 GiB+ | 8 GB GPU | HF files |
| Q5_K_M LiquidAI/LFM2.5-8B-A1B-GGUF | 5.62 GiB | 1 | 6.7 GiB+ | 8 GB GPU | HF files |
| Q8_0 LiquidAI/LFM2.5-8B-A1B-GGUF | 8.39 GiB | 1 | 9.5 GiB+ | 12 GB GPU | HF files |
| FP16 LiquidAI/LFM2.5-8B-A1B-GGUF | 15.78 GiB | 1 | 16.9 GiB+ | 20 GB GPU | HF files |
| MLX 4-bit mixed precision LiquidAI/LFM2.5-8B-A1B-MLX-4bit | 4.50 GiB | 1 | 5.6 GiB+ | 16 GB Mac | HF files |
| MLX 8-bit LiquidAI/LFM2.5-8B-A1B-MLX-8bit | 8.38 GiB | 2 | 9.5 GiB+ | 16 GB Mac | HF files |
Sizes are exact file bytes (text weights; projectors listed separately). “+” = recurrent state or other terms unknown. Smallest fit adds 0.5 GiB spare; Macs use 75% of memory.
Quality
Artificial Analysis Intelligence Index
7.2 AA estimate
Variant: LFM2.5-8B-A1B · observed 2026-10-17 · source
Rank #74 of 75 researched 2026 models.
Arena text rating
Unknown
Scores describe the evaluated hosted variant, not each quantized file. AA and Arena are never averaged.
Reported speeds 0
No reported measurement yet. Submit one.
Community threads
- Tool call sometimes emitted as plain JSON in `content` (missing `<|tool_call_start|>` after `</think>`) — llama.cpp + pi agent Hugging Face · 2026-09-20
- Italian Support Hugging Face · 2026-07-02
Qualitative signals only; titles are not verified claims.