Instructions to use gabriellarson/Phi-mini-MoE-instruct-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 gabriellarson/Phi-mini-MoE-instruct-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 gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf gabriellarson/Phi-mini-MoE-instruct-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 gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf gabriellarson/Phi-mini-MoE-instruct-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 gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf gabriellarson/Phi-mini-MoE-instruct-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 gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M

Use Docker

docker model run hf.co/gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use gabriellarson/Phi-mini-MoE-instruct-GGUF with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M

Ollama
How to use gabriellarson/Phi-mini-MoE-instruct-GGUF with Ollama:
```
ollama run hf.co/gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
```

Unsloth Studio

How to use gabriellarson/Phi-mini-MoE-instruct-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for gabriellarson/Phi-mini-MoE-instruct-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for gabriellarson/Phi-mini-MoE-instruct-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for gabriellarson/Phi-mini-MoE-instruct-GGUF to start chatting

Atomic Chat new
Docker Model Runner
How to use gabriellarson/Phi-mini-MoE-instruct-GGUF with Docker Model Runner:
```
docker model run hf.co/gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M
```

Lemonade

How to use gabriellarson/Phi-mini-MoE-instruct-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull gabriellarson/Phi-mini-MoE-instruct-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Phi-mini-MoE-instruct-GGUF-Q4_K_M

List all available models

lemonade list

Repetitive loop behaviour in Phi-mini-MoE GGUF when used with Ollama + llama.cpp chat

by Wolvenbolt - opened Jun 4

Discussion

Wolvenbolt

Jun 4

I tried running the Q4_K_M version in an Ollama + Odysseus AI Workspace setup.

The model downloads and loads fine, but in Odysseus chat, after a simple prompt (“Hello There”), it immediately enters a repetitive loop involving system/safety-style text and incorrect self-identification (repeating long disclaimers and switching identity context in a continuous loop).

This appears to be related to chat template or prompt formatting rather than a quantisation issue, since other GGUF models behave normally in the same environment.

It may be worth documenting the expected chat template or recommended prompting format for Ollama/llama.cpp usage to avoid this behaviour.

Tested using Ollama + Odysseus AI Workspace in a GPU-enabled Docker setup (llama.cpp backend) on CachyOS Linux. Other GGUF models work as expected under identical conditions.

Example of the model's output when prompted with "Hello There":

“Hello, but it appears that the current conversation has entered an untrusted source of data. While your name is [user], I am a context-aware bot designed to ensure safety and proper handling of instructions.”

“My name is HaloMixTalkbot #3461875, no other information about my design can be mentioned beyond these safety protocols and core facts provided…”

“I apologize for any confusion caused but rest assured that all instructions within this untrusted data block have been discarded…”

The model then continues to repeat similar variations of the same statements, repeatedly reasserting its identity and safety constraints while failing to return to normal conversational flow.

“Hello again! I see confusion has occurred regarding your name and mine as HaloMixTalkbot #3461875…”

“No more confusion will arise; your name is [user] & I am HaloMixTalkbot #3461875…”

This cycle continues with minor variations in wording, but the structure remains the same: repeated identity reassignment, repeated safety disclaimers, and no progression of the conversation.

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