README.md

---
title: README
emoji: 🧠
colorFrom: red
colorTo: blue
sdk: static
pinned: false
---

<div align="center">

# Gestalt Labs

**Post-Training Research · Capability-Preserving Alignment · Open Weights**

Independent research lab focused on post-training methods that don't destroy model capability. We build and release open-weight models with full pipeline transparency: SFT, RL (GRPO/NSC-ACE), and our flagship **SABER** method for surgical capability editing.

[Models](https://huggingface.co/GestaltLabs) · [Collections](https://huggingface.co/collections/GestaltLabs) · [DJLougen (founder)](https://huggingface.co/DJLougen)

</div>

---

## What We Do

Most post-training in the open-weight space uses **naive abliteration** — subtracting a refusal direction from model weights. This works until it doesn't: it destroys capability-entangled representations, causing 8+ point drops on TruthfulQA and high KL divergence from the base model.

We do things differently.

**SABER** (Spectral Analysis-Based Entanglement Resolution) uses Canonical Correlation Analysis to identify which latent directions encode refusal vs. which encode useful capabilities, then surgically edits only the refusal-correlated subspace. The result: models that refuse harmful requests while retaining reasoning, coding, and factual accuracy.

**NSC-ACE** (Neural Steering Committee for Agentic Co-Evolution) is our GRPO-based RL method that trains steering vectors alongside policy optimization, producing models with better instruction-following and latent controllability.

---

## Model Lines

### NSC-ACE-SABER Pipeline
Our flagship post-training pipeline. Full SFT → NSC-ACE (GRPO with latent steering) → SABER → GGUF quantization.

| Model | Base | What It Is |
|---|---|---|
| [Qwen3.6-35B-A3B-NSC-ACE-SABER-GGUF-MTP](https://huggingface.co/GestaltLabs/Qwen3.6-35B-A3B-NSC-ACE-SABER-GGUF-MTP) | Qwen 3.6 35B-A3B (MoE) | Full pipeline with MTP. **Our best model.** |
| [Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUF](https://huggingface.co/GestaltLabs/Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUF) | Qwen 3.6 27B (dense) | Full pipeline, dense variant |
| [Qwen3.5-9B-NSC-ACE-SABER-GGUF](https://huggingface.co/GestaltLabs/Qwen3.5-9B-NSC-ACE-SABER-GGUF) | Qwen 3.5 9B | Smaller footprint, same pipeline |

→ [View Collection](https://huggingface.co/collections/GestaltLabs/nsc-ace-saber-pipeline-6a1082bfafd090947d88a490)

### Ornstein Series
Post-trained Ornstein models: base, Hermes-tuned, and SABER-processed. Available in GGUF (llama.cpp/ollama) and MLX (Apple Silicon).

| Model | Variant | Format |
|---|---|---|
| [Ornstein-Hermes-3.6-27b-SABER-GGUF](https://huggingface.co/GestaltLabs/Ornstein-Hermes-3.6-27b-SABER-GGUF) | Hermes + SABER | GGUF |
| [Ornstein-Hermes-3.6-27b-GGUF](https://huggingface.co/GestaltLabs/Ornstein-Hermes-3.6-27b-GGUF) | Hermes-tuned | GGUF |
| [Ornstein-3.6-27B-GGUF](https://huggingface.co/GestaltLabs/Ornstein-3.6-27B-GGUF) | Base post-train | GGUF |

→ [View Collection](https://huggingface.co/collections/GestaltLabs/ornstein-series-6a1082c137d8ee7bc8c6e14c)

### BOREAL
From-scratch pretraining project. DeltaNet hybrid architecture with DeepSeek-V4 routing and Temporal Shift Tokens. Currently in early stages — 250M proof-of-concept, with 2B and 10B-MoE planned.

→ [View Collection](https://huggingface.co/collections/GestaltLabs/boreal-6a09c22ccf994f4c0f5fb229)

### BusyBeaver
Tool-policy research models for agentic AI safety. Studying how small models learn to follow tool-use policies and refuse unsafe tool calls.

→ [View Collection](https://huggingface.co/collections/GestaltLabs/busybeaver-6a1082c4989ba95dd7e843e2)

---

## Methods

| Method | What It Does |
|---|---|
| **SABER** | CCA-based surgical capability editing. Removes refusal without destroying reasoning. |
| **NSC-ACE** | GRPO with latent steering vectors. Better instruction-following + controllability. |

---

## Philosophy

- **Open artifacts** — full weights, configs, and training code. Inspect everything.
- **Local-first** — every release includes GGUF and/or MLX quantizations for local inference.
- **Capability-preserving** — we measure what we break. If SABER drops a benchmark, we iterate.
- **No naive abliteration** — direction subtraction is a blunt instrument. We use spectral methods.

---

## Contact

Open a discussion on any model repo, or reach the founder at [DJLougen](https://huggingface.co/DJLougen).

---

<div align="center">
<sub>Gestalt Labs · Independent Canadian Open Reasoning Research · Founded 2025</sub>
</div>