Instructions to use AngelSlim/Hy3-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 AngelSlim/Hy3-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 AngelSlim/Hy3-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AngelSlim/Hy3-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 AngelSlim/Hy3-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AngelSlim/Hy3-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 AngelSlim/Hy3-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf AngelSlim/Hy3-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 AngelSlim/Hy3-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf AngelSlim/Hy3-GGUF:Q4_K_M

Use Docker

docker model run hf.co/AngelSlim/Hy3-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use AngelSlim/Hy3-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AngelSlim/Hy3-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": "AngelSlim/Hy3-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/AngelSlim/Hy3-GGUF:Q4_K_M

Ollama
How to use AngelSlim/Hy3-GGUF with Ollama:
```
ollama run hf.co/AngelSlim/Hy3-GGUF:Q4_K_M
```
Unsloth Desktop

How to use AngelSlim/Hy3-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AngelSlim/Hy3-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": "AngelSlim/Hy3-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use AngelSlim/Hy3-GGUF with Docker Model Runner:
```
docker model run hf.co/AngelSlim/Hy3-GGUF:Q4_K_M
```

Lemonade

How to use AngelSlim/Hy3-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull AngelSlim/Hy3-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Hy3-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use AngelSlim/Hy3-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 AngelSlim/Hy3-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 AngelSlim/Hy3-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use AngelSlim/Hy3-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AngelSlim/Hy3-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 "AngelSlim/Hy3-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"

LLAMA.CPP ROCM Vulkan support

by GCBQDevTeam - opened Jul 15

Discussion

GCBQDevTeam

Jul 15

Is it possible to use this with ROCm or Vulkan llama.cpp? Or is it CUDA only?

sjl623

AngelSlim org Jul 15

•

edited Jul 15

Is it possible to use this with ROCm or Vulkan llama.cpp? Or is it CUDA only?

Our patch isn't backend-specific, so in theory it should run on all backend llama.cpp supports. That said, we haven't actually tested it on ROCm/Vulkan ourselves — if anyone in the community is interested, we'd welcome you giving it a try and sharing feedback!

LaikaFramework

Jul 15

Let me try.

LaikaFramework

Jul 15

ROCm can work, but it seems you can't use the latest llama.cpp. I checked the latest code—theoretically it supports hyv3, but the official gguf file fails to load with a tensor error. So I used git to check out this commit: 19bba67c1f4db723c60a0d421aa0788bf4ddc699, merged the official patch, and then it worked.

On my Strix Halo machine, the token speed is around 14 tokens per second.

sjl623

AngelSlim org Jul 15

ROCm can work, but it seems you can't use the latest llama.cpp. I checked the latest code—theoretically it supports hyv3, but the official gguf file fails to load with a tensor error. So I used git to check out this commit: 19bba67c1f4db723c60a0d421aa0788bf4ddc699, merged the official patch, and then it worked.

On my Strix Halo machine, the token speed is around 14 tokens per second.

Thanks for your work! Happy to confirm our patch still work for other backend. As for codebase, yes, you should checkout to the specific commit and apply our patch since hy_v3 arch has not been merged to upstream now.

ffewfqefwsefwfe

Jul 15

ROCm can work, but it seems you can't use the latest llama.cpp. I checked the latest code—theoretically it supports hyv3, but the official gguf file fails to load with a tensor error. So I used git to check out this commit: 19bba67c1f4db723c60a0d421aa0788bf4ddc699, merged the official patch, and then it worked.

On my Strix Halo machine, the token speed is around 14 tokens per second.

Yeah, vanilla llama doesn't seem to work on Strix Halo with image kyuz0/amd-strix-halo-toolboxes:vulkan-radv (latest):

[39967] 0.00.483.411 W load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
[39967] 0.00.528.893 E llama_model_load: error loading model: done_getting_tensors: wrong number of tensors; expected 1278, got 1199
[39967] 0.00.528.901 E llama_model_load_from_file_impl: failed to load model
[39967] 0.00.528.907 E cmn  common_init_: failed to load model '/my-models2/Hy3-GGUF/Hy3-IQ1_M.gguf'
[39967] 0.00.528.912 E srv    load_model: failed to load model, '/my-models2/Hy3-GGUF/Hy3-IQ1_M.gguf'
[39967] 0.00.529.278 E srv  llama_server: exiting due to model loading error

pawarshardul

Jul 16

https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+hy3+is%3Aclosed ----is hy3 is officially supported on llama cpp now?? no need of new patches ???

sjl623

AngelSlim org Jul 16

https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+hy3+is%3Aclosed ----is hy3 is officially supported on llama cpp now?? no need of new patches ???

Thanks for sharing! Will reconvert the GGUF files using the upstream code and simplify the usage instructions soon.

sjl623

AngelSlim org Jul 17

https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+hy3+is%3Aclosed ----is hy3 is officially supported on llama cpp now?? no need of new patches ???

Updated .No patches is needed now. 🚀

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