Kimari Fit Lab — Model Placeholder
⚠️ No weights available yet. Kimari-4B has not been publicly released. This repository serves as a placeholder and compatibility reference.
What This Is
This model card documents the target architecture and compatibility scope for Kimari-4B, a small language model optimized for consumer NVIDIA GPUs (GTX 1060 6GB class).
The companion Kimari Fit Lab Space provides a static GPU compatibility checker — it does not run or download any model.
Compatibility Targets
| Component | Target |
|---|---|
| GPU | NVIDIA GTX 1060 6GB, GTX 1080 8GB |
| VRAM budget | 4–6 GB for inference |
| Quantization | GGUF Q4_K_M |
| Runtime | llama.cpp with CUDA |
| API | OpenAI-compatible endpoint |
Current Status (v0.1.87-alpha)
| Milestone | Status |
|---|---|
| GTX 1060 local runtime | ✅ Validated with TinyLlama test model |
| llama-server CUDA | ✅ Validated |
| OpenAI-compatible endpoint | ✅ Validated |
| HF Jobs GPU smoke | ✅ Completed |
| Private micro SFT | ✅ Completed privately |
| KimariEval Private v1 | ✅ Created, 104 private cases |
| Kimari-4B public release | ❌ Not released |
Links
- GitHub: smouj/kimari-local-ai
- Docs: smouj.github.io/kimari-local-ai
- Space: Kimari Fit Lab
- Org Card: kimari-ai
Gate: BLOCKED — No public weights, adapters, or GGUF files available. No benchmark claims.