RobinsonLabs/Qwen3.5-REAP-212B-A17B-abliterated
Text Generation • 212B • Updated • 305 • 1
AI & ML interests
Self-funded solo AI research lab. We forge open models and run them daily on our own 15-machine, 10-gigabit llama.cpp cluster: abliterated builds, REAP expert-pruned MoE variants, imatrix GGUF ladders, and custom cluster-fit quants. Every release ships with full provenance. Own your models, your memory, your tools.
Recent Activity
View all activity
Organization Card
Robinson Labs
A self-funded solo AI research lab. Local-first. Models forged and run daily on hardware we own.
I run a self-funded solo AI research lab. I have 15+ years of experience developing and designing scalable solutions on the ServiceNow platform. By day I work on enterprise platforms. By night I run a local-first AI operating system on hardware I own, and I forge open models to run on it.
The cluster
It started as a pile of PCs under the desk and a mining-era stack of GPUs. Today it is fifteen machines on a 10-gigabit fabric, two NAS boxes, a custom water loop, and a llama.cpp cluster with over two million tokens of live context across the brain nodes. It serves a 9b classifier lane, multiple 35b chat and agentic models, an 88b fast-chat model, and a 212b production heavy-tier MoE, plus a 30b vision model, an embedding and reranking fleet, and an SDXL image generation box that does four images per pass.
The models here are not drive-by uploads. They are built, quantized, and run daily on that cluster.
What we publish
| Abliterated models | Single-direction weight orthogonalization that relaxes the hard-refusal reflex while keeping harm guardrails intact by design. |
| Abliterated REAP builds | We abliterate community REAP expert-pruned MoE variants (Qwen3.5-212B/262B and the earlier Qwen3.5-122B REAP-30 / REAP-20 cuts) and publish the pair, bf16 base plus the quant ladder. |
| imatrix GGUF quant ladders | Importance-matrix-weighted, Q6_K down to IQ2, so you pick your size and quality tradeoff. |
| Custom cluster-fit quants | Recipes tuned to a specific VRAM and context budget, attention-path precision kept high where it counts and the experts run lighter. |
The models
| Family | bf16 base | GGUF ladder | Architecture |
|---|---|---|---|
| Qwen3.5-REAP-212B-A17B abliterated | base | quants | view graph |
| Qwen3.5-REAP-262B-A17B abliterated | base | quants | view graph |
| Qwen3.6-35B-A3B abliterated | base | quants | view graph |
| Qwen3.5-122B-A10B abliterated | base | quants | view graph |
| Qwen3.5-122B REAP-20 cut, abliterated | base | quants | view graph |
| Qwen3.5-122B REAP-30 cut, abliterated | base | quants | view graph |
The ethos
Every release ships with full provenance and credits the upstream base author. Own your models, your memory, your tools. Lean, efficient, no bloat, runs on anything. The same rule that runs the cluster.
Want a model treated? Open a discussion on our requests board. The stories behind these builds live at robinsonlabs.ai/writing.
models 13
RobinsonLabs/Qwen3.5-REAP-212B-A17B-abliterated-GGUF
Text Generation • 211B • Updated • 3.96k • 1
RobinsonLabs/Qwen3.5-REAP-262B-A17B-abliterated
Text Generation • 262B • Updated • 320 • 1
RobinsonLabs/Qwen3.5-REAP-262B-A17B-abliterated-GGUF
Text Generation • 261B • Updated • 2.04k • 1
RobinsonLabs/Qwen3.5-122B-A10B-REAP-20-abliterated-GGUF
Text Generation • 99B • Updated • 4.89k • 2
RobinsonLabs/Qwen3.5-122B-A10B-REAP-20-abliterated
Text Generation • 99B • Updated • 130 • 2
RobinsonLabs/requests
Updated • 1
RobinsonLabs/Qwen3.6-35B-A3B-abliterated-GGUF
Text Generation • 36B • Updated • 2.33k • 1
RobinsonLabs/Qwen3.6-35B-A3B-abliterated
Text Generation • 36B • Updated • 46 • 1
RobinsonLabs/Qwen3.5-122B-A10B-REAP-30-abliterated-GGUF
Text Generation • 88B • Updated • 3.02k • 1
datasets 0
None public yet