Instructions to use shlbnrj/phi3-kubernetes with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use shlbnrj/phi3-kubernetes with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="shlbnrj/phi3-kubernetes",
	filename="phi3-kubernetes-q4_k_m.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use shlbnrj/phi3-kubernetes with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf shlbnrj/phi3-kubernetes:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf shlbnrj/phi3-kubernetes:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf shlbnrj/phi3-kubernetes:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf shlbnrj/phi3-kubernetes: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 shlbnrj/phi3-kubernetes:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf shlbnrj/phi3-kubernetes: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 shlbnrj/phi3-kubernetes:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf shlbnrj/phi3-kubernetes:Q4_K_M

Use Docker

docker model run hf.co/shlbnrj/phi3-kubernetes:Q4_K_M

LM Studio
Jan
Ollama
How to use shlbnrj/phi3-kubernetes with Ollama:
```
ollama run hf.co/shlbnrj/phi3-kubernetes:Q4_K_M
```

Unsloth Studio new

How to use shlbnrj/phi3-kubernetes 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 shlbnrj/phi3-kubernetes 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 shlbnrj/phi3-kubernetes to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for shlbnrj/phi3-kubernetes to start chatting

Docker Model Runner
How to use shlbnrj/phi3-kubernetes with Docker Model Runner:
```
docker model run hf.co/shlbnrj/phi3-kubernetes:Q4_K_M
```

Lemonade

How to use shlbnrj/phi3-kubernetes with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull shlbnrj/phi3-kubernetes:Q4_K_M

Run and chat with the model

lemonade run user.phi3-kubernetes-Q4_K_M

List all available models

lemonade list

phi3-kubernetes

Phi-3-mini fine-tuned on 300 Kubernetes Q&A samples from Stack Overflow using QLoRA. Built as part of an end-to-end Local Research Assistant pipeline. See the project repo for full context, including dataset preparation, evaluation results, inference server, and CI/CD setup.

Files

File	Size	Purpose
`phi3_kubernetes_lora.zip`	106 MB	LoRA adapter weights (rank=16, alpha=32)
`phi3-kubernetes-q4_k_m.gguf`	2.16 GB	Merged GGUF, 4-bit quant — recommended deployment
`phi3-kubernetes-q8_0.gguf`	not hosted	Benchmarked locally; regenerable with `llama-quantize` from the LoRA adapter

The Q8_0 quantization was benchmarked during development (see eval section below) but is not hosted here to keep the repo small. The Q4_K_M is the deployment-recommended variant: ~3.7× faster throughput at the same VRAM ceiling.

Training

Hyperparameter	Value
Base model	microsoft/Phi-3-mini-4k-instruct
Fine-tune method	QLoRA (Unsloth)
LoRA rank	16
LoRA alpha	32
Learning rate	2e-4 (cosine schedule)
Batch size	8 (effective 16 via grad accumulation)
Epochs	3
Hardware	Colab T4 (16 GB VRAM)
Dataset	mcipriano/stackoverflow-kubernetes-questions (300 samples filtered)

Evaluation

Tested on 30 held-out K8s Q&A samples against the base phi3:mini model:

Metric	Base phi3:mini	Fine-tuned phi3-kubernetes	Δ
ROUGE-L	0.1382	0.1622	+17.4%
Avg latency	18,104 ms	9,272 ms	2× faster

Quantization benchmark (RTX 3050 Laptop, 4 GB VRAM)

Quant	TTFT (ms)	Throughput (tok/s)	Peak VRAM
Q4_K_M	2,692	63.6	3.4 GB
Q8_0	2,798	17.3	3.6 GB

Q4_K_M is ~3.7× faster at the same VRAM ceiling — chosen as the deployment default.

Usage with Ollama

ollama run hf.co/shlbnrj/phi3-kubernetes:Q4_K_M

Or manually:

wget https://huggingface.co/shlbnrj/phi3-kubernetes/resolve/main/phi3-kubernetes-q4_k_m.gguf

cat > Modelfile <<'EOF'
FROM ./phi3-kubernetes-q4_k_m.gguf
PARAMETER temperature 0.4
PARAMETER num_ctx 4096
EOF

ollama create phi3-kubernetes -f Modelfile
ollama run phi3-kubernetes "What is a Pod?"

License

MIT. The base model (Phi-3) is under the Microsoft Research License.

Limitations

This is a small (3.8B parameter) model. It works well for direct K8s knowledge questions but is unreliable for multi-step tool-use scenarios (e.g., chaining a search with a Python computation). See the project's NOTES.md for detailed failure-mode analysis.

Downloads last month: 21

GGUF

Model size

4B params

Architecture

llama

Hardware compatibility

4-bit

Inference Providers NEW

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

Model tree for shlbnrj/phi3-kubernetes

Base model

microsoft/Phi-3-mini-4k-instruct

Adapter

(839)

this model