Instructions to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF 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 SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

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 SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

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 SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

Use Docker

docker model run hf.co/SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with Ollama:
```
ollama run hf.co/SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
```
Unsloth Desktop

How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

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": "SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with Docker Model Runner:
```
docker model run hf.co/SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M
```

Lemonade

How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.GLM-5.3-Flash-CYBERSECURITY-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

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 SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M

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 "SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF:Q4_K_M" \
  --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 CYBERSECURITY — GGUF Dynamic Quant Ladder (MTP inside)

Full-size dynamic GGUF ladder of dealignai/GLM-5.3-Flash-CYBERSECURITY-W4A16 (~321B-param MoE, cyber-tuned), rebuilt from a local Q8_0 master with the MTP draft head preserved and verified — then requantized with an imatrix-guided tensor budget that spends bits where intelligence actually lives.

Every *-MTP.gguf keeps the nextn draft head → run with --spec-type draft-mtp for self-speculation. No separate draft model needed.

🪜 The Ladder

File	Size	~bpw	Down-proj	MTP head	Min realistic rig
`GLM-5.3-Flash-CYBER-Q4_K_M-MTP.gguf`	182.1 GB	4.5	q5_K	q8_0	2× RTX PRO 6000 (192 GB, tight) · 8× 4090
`GLM-5.3-Flash-CYBER-IQ4_XS-MTP.gguf`	163.9 GB	4.1	iq4_nl	q8_0	2× PRO 6000 (comfy) · 8× 3090
`GLM-5.3-Flash-CYBER-IQ3_M-MTP.gguf`	144.2 GB	3.6	iq4_nl	q8_0	2× PRO 6000 · 8× 3090
`GLM-5.3-Flash-CYBER-IQ2_M-MTP.gguf`	113.5 GB	2.8	iq3_s	blk.45/eh_proj/enorm q8_0	4× 5090 · 2× A100-80 · 8× 3090
`GLM-5.3-Flash-CYBER-IQ1_M-MTP.gguf`	~90 GB	2.2	iq2_xs	blk.45/eh_proj/enorm q8_0	1× PRO 6000 (96 GB!) · 4× 3090

Legacy (no MTP, superseded): IQ3_M.gguf 140.9 GB, IQ2_M.gguf 107.6 GB.

🧠 The Tensor Budget (why these run smart at low bpw)

Uniform quants waste bits. This ladder doesn't — every quant is requantized from the Q8 master with a fresh imatrix (calib-flash-smart, 32 chunks, security-weighted corpus) and a fixed protection map:

Tensor	Type	Why
`token_embd`	Q8_0	every token passes through
`output`	Q6_K	logits live here
attention (`attn_q/k/v/output`, `attn_q_b/kv_b/k_b`)	Q5_K / Q6_K	tiny param share, outsized damage
`ssm`	Q5_K	hybrid dynamics layers
`ffn_gate_inp` (router)	Q6_K	one wrong bit misroutes every expert
shared experts (`ffn_*_shexp`)	Q5_K	always-on, zero redundancy
`ffn_down_exps`	q5_K → iq2_xs (by tier)	most sensitive expert tensor
gate/up experts	body ftype (Q4_K_M → IQ1_M)	the bulk — hammered
nextn / MTP	Q8_0	draft quality = acceptance rate

At the IQ1_M rung the body is IQ1_M with the router still at Q6_K and the down-proj cushioned at iq2_xs — ~2.2 bpw that behaves like a carefree 2.5.

🚀 Serving (llama.cpp master)

llama-server \
  -m GLM-5.3-Flash-CYBER-IQ4_XS-MTP.gguf \
  -ngl 99 --flash-attn on --cache-type-k q8_0 --cache-type-v q8_0 \
  --spec-type draft-mtp --spec-draft-n-max 2 \
  --jinja --ctx-size 32768

MTP verified post-convert (nextn tensors present, checked programmatically before any quant ran)
--reasoning-budget-style chat template args supported via --jinja
Q8_0 KV cache recommended; fits within the table's rig guidance

📜 Provenance

Source: dealignai/GLM-5.3-Flash-CYBERSECURITY-W4A16 (4.25 bpw crack pack, 181.6 GB)
Master: converted to Q8_0 (9 × 40 GB shards, 341 GB) — MTP tensors verified
imatrix: fresh, CPU, 32 chunks over the smart cyber corpus
Ladder: dynamic requants per the table above, each uploaded then locally cleaned
Full chain (download → convert → verify → imatrix → 5 quants) ran unattended in ~13 h

🙏 Credits

dealignai — the W4A16 cyber-tune this ladder is built from
The llama.cpp imatrix / i-quant ecosystem
GLM-5.3-Flash is MoE: experts forgive low bits, routers don't — quantize accordingly

⚠️ Responsible Use

Cybersecurity-tuned and refusal-leaned: intended for defense, red-teaming, security research, and evaluation. You are responsible for what you do with it. Don't do crimes.

Downloads last month: 8,228

GGUF

Model size

313B params

Architecture

glm5-next

Hardware compatibility

1-bit

2-bit

3-bit

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF

Base model

zai-org/GLM-5.3-Flash

Quantized

dealignai/GLM-5.3-Flash-CYBERSECURITY-W4A16

Quantized

(1)

this model

SyndicateLabs
/

GLM-5.3-Flash-CYBERSECURITY-GGUF

GLM-5.3-Flash CYBERSECURITY — GGUF Dynamic Quant Ladder (MTP inside)

🪜 The Ladder

🧠 The Tensor Budget (why these run smart at low bpw)

🚀 Serving (llama.cpp master)

📜 Provenance

🙏 Credits

⚠️ Responsible Use

Model tree for SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF

Space using SyndicateLabs/GLM-5.3-Flash-CYBERSECURITY-GGUF 1