Fikra Model Series
Highly optimized ternary-weight models designed for consumer hardware. State-of-the-art reasoning with 1.58-bit quantization.
| Model Name | Parameters | Quantization | VRAM/RAM | Context | Action |
|---|---|---|---|---|---|
|
Fikra-1B-Nano
v0.2 • Release Candidate
|
1.2B | w1.58 (Ternary) | ~2.1 GB | 4k | HuggingFace |
|
Fikra-3B-Edge
In Training
|
3.0B | w1.58 | ~4.5 GB | 8k | Coming Q2 |
Updated: Feb 2026
SHA256 Verified
Installation
The Lacesse Python SDK handles model downloading, caching, and inference automatically.
$ pip install fikra
Hardware Requirements
- Apple Silicon: M1, M2, M3 (Native Metal support)
- Linux/x86: AVX2 compatible CPU (Intel i5 8th gen+)
- Raspberry Pi: Pi 5 (8GB) Recommended