Instructions to use patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M

Use Docker

docker model run hf.co/patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M

Ollama
How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with Ollama:
```
ollama run hf.co/patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
```
Unsloth Desktop

How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf patrickbdevaney/GLM-5.3-Flash-REAP50-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": "patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with Docker Model Runner:
```
docker model run hf.co/patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M
```

Lemonade

How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.GLM-5.3-Flash-REAP50-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-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 patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use patrickbdevaney/GLM-5.3-Flash-REAP50-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf patrickbdevaney/GLM-5.3-Flash-REAP50-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 "patrickbdevaney/GLM-5.3-Flash-REAP50-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"

Seems to be OK in first short tests. The most accessible and workable of GLM 5.3 Flash deployments

by iamtanmay - opened Sep 26

Discussion

iamtanmay

Sep 26

Unlike your MiMoV2.6 Flash REAP which bugged during reasoning (max), this model seems clean and stable at first glance. Since I am getting 5-7 tok/s currently, I haven't done longer tests yet.

I will conduct longer tests

iamtanmay

30 days ago

I did another day of testing... the prune itself is really good, it doesn't seem to cause problems with the underlying model

However, the model itself is rather unsuited for a homelab. I can run Deepseek v4.1 Flash at 9-10 tok/s over tens of thousands of token context on 2 32GB GPUs, Qwen Next on a single 32GB GPU at 21 tok/s

GLM starts at 7 tok/s, but by 20k context its down to 1.4 tok/s... on 3 32GB GPUs

The cloud version of GLM 5.3 flash is the most capable model on the market today, so I was really hoping to get it working locally.

Its a real shame, but I hope they learn from Qwen and Deepseek to improve their efficiency

BingoBird

9 days ago

I did another day of testing... the prune itself is really good, it doesn't seem to cause problems with the underlying model

However, the model itself is rather unsuited for a homelab. I can run Deepseek v4.1 Flash at 9-10 tok/s over tens of thousands of token context on 2 32GB GPUs, Qwen Next on a single 32GB GPU at 21 tok/s

GLM starts at 7 tok/s, but by 20k context its down to 1.4 tok/s... on 3 32GB GPUs

The cloud version of GLM 5.3 flash is the most capable model on the market today, so I was really hoping to get it working locally.

Its a real shame, but I hope they learn from Qwen and Deepseek to improve their efficiency.

https://github.com/neurall/llama.cpp

On 1x 3090 + 128GB ddr5 2ch i see 11-12t/s at high ctx, using MoE caching version of llama.cpp
Cache hits quite poor at 40% unless i add another 24GB GPU.

iamtanmay

9 days ago

Yeah I noticed other people got much higher tok/s than me on similar hardware. I have to rerun and check what could be the bottleneck. The model itself is top quality, better than Qwen/Deepseek and still a small enough size to be feasible to run locally

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment