Aleph Alpha Kolibri-1
Files and memory at 8K context
| Quant | File | Files | + KV + runtime | Smallest fit | Source |
|---|---|---|---|---|---|
| Q4_K_M Hob-forge/Kolibri-1-GGUF | 44.20 GiB | 1 | 45.4 GiB+ | 48 GB GPU | HF files |
| Q5_K_M Hob-forge/Kolibri-1-GGUF | 51.77 GiB | 2 | 53.0 GiB+ | 64 GB GPU | HF files |
| Q8_0 Hob-forge/Kolibri-1-GGUF | 77.43 GiB | 2 | 78.6 GiB+ | 80 GB GPU | HF files |
| MLX 4-bit mixed precision velaia/Kolibri-1-MLX-4bit | 41.29 GiB | 9 | 42.5 GiB+ | 64 GB Mac | HF files |
No verified file: FP16, BF16, MLX 8-bit. 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
Unknown
Arena text rating
Unknown
Scores describe the evaluated hosted variant, not each quantized file. AA and Arena are never averaged.
Reported speeds 6
| Hardware | Quant | Backend | Prompt t/s | Gen t/s | Date | Source |
|---|---|---|---|---|---|---|
| AMD Ryzen 7 7800X3D + 128 GB DDR5-6400, CPU only (RTX 5070 hidden) | Q4_K_M | llama.cpp CPU | 83 | 12.3 | 2026-10-07 | huggingface.co |
| AMD Ryzen 7 7800X3D + 128 GB DDR5-6400 + RTX 5070 12 GB (attention/shared expert on GPU, routed experts in RAM) | Q4_K_M | llama.cpp CUDA | 723 | 18 | 2026-10-07 | huggingface.co |
| AMD Ryzen 7 7800X3D + 128 GB DDR5-6400 + RTX 5070 12 GB (attention/shared expert on GPU, routed experts in RAM) | Q4_K_M | llama.cpp CUDA | 696 | 11.7 | 2026-10-07 | huggingface.co |
| Apple M4 Max (40-core GPU), 64 GB unified memory | MLX 4-bit (community conversion) | MLX | 900 | 93.1 | 2026-10-06 | omlx.ai |
| Apple M5 Max (40-core GPU), 128 GB unified memory | MLX 4-bit (community conversion) | MLX | 2,365 | 128 | 2026-10-05 | omlx.ai |
| Apple M5 Max (40-core GPU), 128 GB unified memory | MLX 4-bit (community conversion) | MLX | 1,739 | 94.6 | 2026-10-05 | omlx.ai |
Reported by the linked sources, not reproduced by LocalFit. All rows in the speed database