Instructions to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Pinkstack/PARM-Phi-3.5-mini-o1-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Pinkstack/PARM-Phi-3.5-mini-o1-GGUF", dtype="auto")

llama-cpp-python

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Pinkstack/PARM-Phi-3.5-mini-o1-GGUF",
	filename="Pinkstack-model-o1.Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Pinkstack/PARM-Phi-3.5-mini-o1-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 Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Pinkstack/PARM-Phi-3.5-mini-o1-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 Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Pinkstack/PARM-Phi-3.5-mini-o1-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": "Pinkstack/PARM-Phi-3.5-mini-o1-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M

SGLang

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Pinkstack/PARM-Phi-3.5-mini-o1-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Pinkstack/PARM-Phi-3.5-mini-o1-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Pinkstack/PARM-Phi-3.5-mini-o1-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Pinkstack/PARM-Phi-3.5-mini-o1-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with Ollama:
```
ollama run hf.co/Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
```

Unsloth Studio new

How to use Pinkstack/PARM-Phi-3.5-mini-o1-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 Pinkstack/PARM-Phi-3.5-mini-o1-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 Pinkstack/PARM-Phi-3.5-mini-o1-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Pinkstack/PARM-Phi-3.5-mini-o1-GGUF to start chatting

Docker Model Runner
How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with Docker Model Runner:
```
docker model run hf.co/Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M
```

Lemonade

How to use Pinkstack/PARM-Phi-3.5-mini-o1-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Pinkstack/PARM-Phi-3.5-mini-o1-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.PARM-Phi-3.5-mini-o1-GGUF-Q4_K_M

List all available models

lemonade list

PARM V2

🧀 Which quant is right for you?

Q4: This model should be used for low end devices like phones or older laptops due to its compact size, quality is okay but fully usable.
Q5: This model should be used on devices which are decently powerful, eg; gtx 1650 gpu's or better for quick responses. Quality is a bit better than Q4.
Q8: This model is the best one we offer, should be used on high end devices such as rtx 3070 gpu's or better, respones are very high quality and it's better at reasoning than Q5, Q4.

Things you should be aware of when using PARM models (Pinkstack Accuracy Reasoning Models) 🧀

This PARM is based on Phi 3.5 mini which has gotten extra training parameters so it would have similar outputs to O.1 Mini, We trained with this dataset.

To use this model, you must use a service which supports the GGUF file format. Additionaly, this is the Prompt Template, it uses the Phi-3 template.

{{ if .System }}<|system|> {{ .System }}<|end|> {{ end }}{{ if .Prompt }}<|user|> {{ .Prompt }}<|end|> {{ end }}<|assistant|> {{ .Response }}<|end|>

Or if you are using an anti prompt: <|end|><|assistant|>

Highly recommended to be used with a system prompt.

This model has been tested inside of:

Msty: with 8,192 Max token output and 32,000 Context. (RTX 3080, q8 model) Very high quality responses
Ollama: with 1,000 Max token output and 1,000 Context. (Qualcomm Snapdragon 8 Gen 2, q5 model) High quality responses
Transformers: with 4,096 Max token output and 2,048 Context. (Nvidia tesla T4, q4 model) Medium quality but useable responses

Extra information

Developed by: Pinkstack
License: MIT
Finetuned from model : unsloth/phi-3.5-mini-instruct-bnb-4bit

This llama model was trained with Unsloth and Huggingface's TRL library.

Used this model? Don't forget to leave a like :)

Downloads last month: 100

GGUF

Model size

4B params

Architecture

llama

Hardware compatibility

4-bit

5-bit

8-bit

Collection including Pinkstack/PARM-Phi-3.5-mini-o1-GGUF

Parm V1

Collection

First gen of the Pinkstack advanced reasoning models. Medium quality but better than the original models they are based on. • 3 items • Updated Sep 22, 2025 • 2