52 lines
1.7 KiB
Markdown
52 lines
1.7 KiB
Markdown
# Organ Architecture
|
|
|
|
**Neural network surgery and checkpoint engineering**
|
|
|
|
[Inference-X](https://inference-x.com) · [Community](https://git.inference-x.com/inference-x-community) · [Organ Store](https://git.inference-x.com/inference-x-community/organ-store)
|
|
|
|
---
|
|
|
|
Organ Architecture provides tools to analyze, modify, and recombine AI model components. Visualize attention heads, extract MLP layers, transplant capabilities between models — all locally.
|
|
|
|
## What You Can Do
|
|
|
|
- **Visualize** — Inspect layer activations, attention patterns, embedding geometry
|
|
- **Extract** — Pull specific layers, heads, or modules from any GGUF model
|
|
- **Transplant** — Merge capabilities from multiple models
|
|
- **Prune** — Remove redundant layers to create smaller, faster models
|
|
- **Quantize** — Convert between precisions (F32 → F16 → Q8 → Q4 → Q2)
|
|
|
|
## Supported Model Families (Feb 2026)
|
|
|
|
Llama 3.x · Qwen 2.5 · DeepSeek R1 · Mistral · Gemma 2 · Phi-4 · Command-R
|
|
|
|
## Quick Start
|
|
|
|
```bash
|
|
git clone https://git.inference-x.com/inference-x/organ-architecture
|
|
cd organ-architecture && make
|
|
|
|
# Analyze a model
|
|
./organ analyze model.gguf
|
|
|
|
# Visualize attention patterns
|
|
./organ visualize model.gguf --layer 16 --head 0
|
|
|
|
# Extract layers 0-20
|
|
./organ extract model.gguf --layers 0-20 --output model_small.gguf
|
|
|
|
# Merge two models (50/50)
|
|
./organ merge base.gguf specialist.gguf --ratio 0.5 --output merged.gguf
|
|
|
|
# Quantize F16 → Q4_K_M
|
|
./organ quantize model.f16.gguf --target Q4_K_M
|
|
```
|
|
|
|
## Organ Store Integration
|
|
|
|
Share and download model organs at [git.inference-x.com/inference-x-community/organ-store](https://git.inference-x.com/inference-x-community/organ-store).
|
|
|
|
---
|
|
|
|
[inference-x.com](https://inference-x.com) · BSL 1.1
|