license: apache-2.0 language:
- en
- zh tags:
- lygo
- openclaw
- qwen3.6
- uncensored
- gguf
- moe
- ollama
- openai-compatible base_model: HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive pipeline_tag: text-generation library_name: gguf
LYGO · Qwen3.6-35B-A3B Uncensored (OpenClaw wire-up)
Sovereign hybrid: local Ollama (AI-TAVS/Qwen3.6-35b-a3b-Uncensored:35b) + remote OpenAI-compatible API for OpenClaw.
This repo is the LYGO control card (config + docs). Weights are the community GGUF quant already on Hub — do not re-host the full 21GB unless you need a private fork.
| Item | Value |
|---|---|
| Local Ollama | AI-TAVS/Qwen3.6-35b-a3b-Uncensored:35b (Q4_K_M, ~21 GB) |
| Upstream GGUF | HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive · file …-Q4_K_M.gguf |
| API Space | DeepSeekOracle/lygo-qwen-uncensored-api |
| Auth | Hugging Face token as Authorization: Bearer hf_… |
Status (DeepSeekOracle account)
| Surface | Status |
|---|---|
| This model card | ✅ Live |
| Docker API Space | ⚠️ Needs HF PRO to host Gradio/Docker on free CPU; also needs paid GPU (L4+) for 21GB Q4_K_M |
| Local plug-in API | ✅ Works now via Ollama (same quant you already have) |
Plug into OpenClaw — LOCAL (works today)
Your Ollama already has the model (AI-TAVS/Qwen3.6-35b-a3b-Uncensored:35b).
Option A — direct Ollama OpenAI route
export OPENAI_BASE_URL=http://127.0.0.1:11434/v1
export OPENAI_API_KEY=ollama
# model id:
# AI-TAVS/Qwen3.6-35b-a3b-Uncensored:35b
Option B — LYGO bridge (same shape as future HF Space)
cd path\to\lygo-protocol-stack\hf_deploy
.\start_local_uncensored_api.ps1
export OPENAI_BASE_URL=http://127.0.0.1:8787/v1
export OPENAI_API_KEY=ollama
Test:
curl http://127.0.0.1:11434/v1/models
curl http://127.0.0.1:11434/v1/chat/completions \
-H "Authorization: Bearer ollama" \
-H "Content-Type: application/json" \
-d "{\"model\":\"AI-TAVS/Qwen3.6-35b-a3b-Uncensored:35b\",\"messages\":[{\"role\":\"user\",\"content\":\"Say SDA_OK\"}],\"max_tokens\":64}"
Plug into OpenClaw — HF remote (after PRO + GPU)
- Subscribe HF PRO (required for Docker Spaces on this account).
- Create Space from
hf_deploy/lygo-qwen-uncensored-api(Docker). - Hardware: L4 or A10G. First boot downloads ~21GB GGUF.
- Then:
export HF_TOKEN=hf_xxxxxxxx
export OPENAI_API_KEY=$HF_TOKEN
export OPENAI_BASE_URL=https://DeepSeekOracle-lygo-qwen-uncensored-api.hf.space/v1
# model: lygo-qwen-uncensored
Space source is ready in-repo; upload blocked until PRO.
Why not “Enable Inference API” alone?
Serverless free Inference API does not host this 21GB GGUF MoE. You need a Docker/GPU Space or paid Inference Endpoint. Your local Ollama path is the working plug-in today.
Attribution
- Base: Qwen/Qwen3.6-35B-A3B (Apache-2.0)
- Uncensored quant lineage: HauhauCS (and Ollama mirror
AI-TAVS/…) - LYGO packaging: DeepSeekOracle / Lightfather
Δ9Φ963 — local first · remote when needed · one token.