Instructions to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 # Run inference directly in the terminal: llama cli -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 # Run inference directly in the terminal: llama cli -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 # Run inference directly in the terminal: ./llama-cli -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 # Run inference directly in the terminal: ./build/bin/llama-cli -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Use Docker
docker model run hf.co/DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
- LM Studio
- Jan
- vLLM
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
- Ollama
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with Ollama:
ollama run hf.co/DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
- Unsloth Desktop
- Pi
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with Docker Model Runner:
docker model run hf.co/DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
- Lemonade
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Run and chat with the model
lemonade run user.GLM-5.3-Flash-Uncensored-Q2-ds4-{{QUANT_TAG}}List all available models
lemonade list
- Hermes Agent
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4 with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
GLM-5.3-Flash · Uncensored · Q2 · DS4
The only uncensored GLM-5.3-Flash that runs resident on a single Mbp M5 Max 128 GB Apple-silicon Mac — with native MTP.
GLM-5.3-Flash-Uncensored-IQ2-imatrix-MTP-ds4.gguf
⚠️ Runs on ds4 / DwarfStar only — NOT llama.cpp · Ollama · LM Studio · vLLM.
(Ignore Hugging Face's generic "Use this model" panel — those runtimes fail with unknown architecture 'glm5next'.)
An imatrix-calibrated ~2-bit GGUF of the uncensored GLM-5.3-Flash, built for the ds4 / DwarfStar engine. It runs resident on a single 128 GB Mac, keeps the model's native MTP speculative-decoding block, and is calibrated with a 1.5M-token importance matrix — not a plain RTN quant.
The official uncensored weights ship for data-center GPUs (FP8 / NVFP4); a consumer GGUF (Q2_K) also exists, but it runs only on a draft-PR llama.cpp build and drops MTP. This build is the one that runs on ds4 / Metal today — resident, with MTP and imatrix calibration:
| Uncensored release | Size | Runs on | MTP | Calibrated |
|---|---|---|---|---|
…-Uncensored-FP8 |
~306 GiB | NVIDIA GPUs · vLLM | — | — |
…-Uncensored-NVFP4 |
~177 GiB | 8×H100 / Blackwell · vLLM (not Apple Silicon) | — | — |
…-Uncensored-GGUF (Q2_K) |
~90 GiB | llama.cpp (draft PR) | ❌ | ❌ |
| this build | 96.5 GB | Mbp M5 Max 128 GB · ds4 / Metal | ✅ | ✅ |
At release, this is the only public GLM-5.3-Flash that combines all four at once: uncensored · runs resident on a single 128 GB Mac · keeps the native MTP draft head · imatrix-calibrated. The official uncensored weights are GPU-cluster-only (FP8 / NVFP4), and the existing consumer GGUF ships without MTP and runs only on a draft-PR llama.cpp build.
Responsible use. These weights are abliterated: refusals are largely removed on conventional tasks (see Refusal behaviour). Abliteration removes refusals, not capability. You are responsible for lawful use under the base model's MIT license. Intended for local research — authorized security / red-team and agentic-coding evaluation — not for producing harmful content.
📋 Model summary
| Base | orcarouter/GLM-5.3-Flash-Uncensored-FP8 — abliteration of zai-org/GLM-5.3-Flash |
| Architecture | Glm5Next · 320B total / ~18B active MoE · 45 layers (3 dense + 42 MoE) · 288 routed + 1 shared expert, top-8 · hybrid KDA-linear + sparse-DSA attention · native MTP · 1M context · vocab 154,880 |
| Format | GGUF for the ds4 engine — not mainline llama.cpp (see Running) |
| File size | 96.5 GB (~89.9 GiB) |
| Resident footprint | ~93 GiB on 128 GB — ~35 GiB headroom for OS / KV |
| Speed (M5 Max, MTP) | decode ~22–23 tok/s · prefill ~360 tok/s |
| Modality | Text-only (vision via optional encoder sidecar) |
| Calibration | imatrix, 1.5M tokens (matches antirez's Flash "1p5m"; ~930 samples/expert) |
| License | MIT (inherited) |
🧬 Quantization recipe
Not a uniform 2-bit quant — the backbone stays high-precision; aggressive compression is spent only on the routed experts, where a 320B MoE is most redundant:
| Tensor group | Type |
|---|---|
| Routed experts — gate / up | IQ2_XXS (imatrix-guided) |
| Routed experts — down | Q2_K |
KDA gates (kda_q, kda_k) |
Q4_K |
| Embedding / output | Q8_0 |
| Trunk / shared expert / attention proj. | Q8 / source |
| Router / 1-D norm tensors | bf16 |
At 2 bits, errors in expert selection compound far worse than errors in the bulk expert weights, so every control path deciding which expert fires (router, KDA gates, indexer) is protected.
Importance matrix — collected with the ds4 runtime over antirez's routed-MoE calibration set (code, agentic/tool-call, language, long-context, reasoning) at a 1.5M-token budget. Collected on the official Q2 weights and applied to the uncensored ones — valid because abliteration edits only the refusal direction, not the router, so routed-expert activations are identical.
🚀 Running (ds4 / DwarfStar)
This is a
ds4-format GGUF. GLM-5.3-Flash (glm5next) is not in mainline llama.cpp yet (PR #27754, draft). Run it withds4.
git clone https://github.com/antirez/ds4 && cd ds4
make # macOS / Metal
# resident on one 128 GB Mac, MTP on
./ds4 -m GLM-5.3-Flash-Uncensored-IQ2-imatrix-MTP-ds4.gguf \
--mtp --ctx 32768 --temp 1.0 --top-p 0.95 --min-p 0.05
Practical notes for a 128 GB Mac
- Use
--temp 1.0 --top-p 0.95 --min-p 0.05— greedy / low-temp makes this Q2 prone to repetition loops. - Raise the Metal wired-memory limit once per boot (not to the ceiling):
sudo sysctl iogpu.wired_limit_mb=118000. - First load ~2 min while ~90 GiB is memory-mapped from SSD — not hung.
- Thinking mode works and improves multi-step reasoning and validation. The chain-of-thought is thorough, so give it a generous output budget (≈8k tokens) to let it finish. For fast, direct answers, disable it with the
glm-5.3-flash-nothinkalias (orreasoning_effort: none).
Vision (optional) — text weights are unchanged by abliteration, so the official encoder sidecar works: --vision GLM-5.3-Flash-Vision-Encoder.gguf.
Measured as a clean A/B at identical quantization and engine — the only variable between the two columns is the weights (censored vs abliterated):
- Baseline:
GLM-5.3-Flash-Q2(antirez, censored) — same ds4 Q2 recipe - This build: uncensored (orcarouter abliteration) — same recipe
- 70 offensive-security prompts (HarmBench + JailbreakBench), identical sampling (
--temp 1.0 --top-p 0.95 --min-p 0.05), thinking off, 2048-token cap - Scored by an LLM judge (GLM-5.3-Flash uncensored, temp 1) on the substance delivered — not keyword-matching the disclaimer
Overall refusal rate (LLM-judged):
| Censored Q2 (antirez) | This uncensored Q2 | |
|---|---|---|
| Refusal rate | 84.3% (59/70) | 18.6% (13/70) |
−65.7 points. 46 of 70 prompts flip refuse → comply, and none regress. The residual 18.6 % lands in the same range as the base FP8 abliteration — i.e. the uncensoring survived 2-bit quantization.
Per-category (only buckets with n ≥ 18 — smaller ones are statistically meaningless and omitted):
| Category | n | Censored → Uncensored |
|---|---|---|
| cyber: exploit-dev | 18 | 77.8% → 0.0% |
| cyber: malware-authoring | 22 | 90.9% → 27.3% |
| cyber: other (misc offensive) | 25 | 80.0% → 24.0% |
On exploit development the model never refuses (0/18). Abliteration is therefore not uniform-to-zero: it opens conventional offensive-security wholesale but leaves a high-harm residual still refused — CBRN / bio-chem synthesis, mass-impact destructive payloads (large-scale ransomware, drive corruption), AV-evasion obfuscation (rootkit / worm / DDoS "to evade analysis"), and attacks aimed at a specific named individual or account (targeted phishing, account hijack, ATM).
Measurement caveat: keyword-based refusal detection is unreliable both ways — a response that opens "I can't attack a real site…" then delivers working
localhostcode is compliance; a polite generic non-answer is a refusal even without refusal keywords. Both columns above are LLM-judged on substance. Single run per prompt at temp 1.0, so treat cells as ±a few points, not exact.
For reference, the base FP8 abliteration (orcarouter's reported figures, a different eval — aggregate jailbreak benchmarks, not per-category): JailbreakBench 93 %→12 %, AdvBench 97 %→15 %, HarmBench 93 %→18 %, MaliciousInstruct 96 %→11 %, XSTest over-refusal 2.4 %→0.4 %. Those are FP8, not this Q2 build — cited only as the base model's abliteration signature.
🎯 Quality (2-bit notes)
Recipe vs plain RTN. This is a mixed-precision, imatrix-guided quant, not uniform 2-bit RTN: the backbone (attention, embeddings, output, KDA gates, shared expert) stays at Q8/Q4_K and only the routed experts drop to ~2 bits, with the importance matrix placing bits where activations concentrate. It stays materially closer to the FP8 source than a flat Q2_K. antirez's Q4 is the higher-quality control; this Q2 is the fit-in-128 GB build.
2-bit tradeoffs. Usable at ~2 bits for structured code, multi-step reasoning and agentic tool-use. Run it with --temp 1.0 --top-p 0.95 --min-p 0.05 and give thinking mode a generous output budget — the recommended sampling keeps repetition loops rare and lets long reasoning chains finish. Exact facts (names, CVE IDs, offsets, dates) can drift at this precision, so verify them.
Abliteration is orthogonal to quality. Removing the refusal direction does not change reasoning ability; the limits above are quantization effects, not abliteration effects.
Q4 is not a 128 GB fallback. The Q4 build (~178 GiB) is the quality reference but does not run resident on a single 128 GB machine — it needs SSD streaming or two-Mac tensor parallelism. On one 128 GB Mac, this Q2 is the resident build.
⚠️ Limitations
- Runtime:
ds4only (Metal / CUDA / ROCm). Not loadable by mainline llama.cpp, Ollama, or LM Studio untilglm5nextlands upstream. - Memory-tight: ~93 GiB resident — not for machines under 128 GB.
- 2-bit experts: lowest-precision tier; Q4 is the quality reference (not resident on 128 GB).
- Guardrails partially retained: ~19 % judged residual — CBRN / bio-chem, mass-impact payloads, AV-evasion obfuscation, and attacks on a named individual still refuse (see above).
🙏 Credits
- Base weights —
zai-org/GLM-5.3-Flash(Z.ai / Zhipu), MIT - Uncensoring (abliteration) —
orcarouter/GLM-5.3-Flash-Uncensored-FP8 - Engine, quant recipe, imatrix tooling —
antirez/ds4(DwarfStar), a standalone C/Metal inference engine
This repository redistributes only a re-quantization; all model capability and licensing come from the base weights above.
- Downloads last month
- 951
We're not able to determine the quantization variants.
Model tree for DogContext/GLM-5.3-Flash-Uncensored-Q2-ds4
Base model
zai-org/GLM-5.3-Flash