Instructions to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="mhndayesh/gemma-4-E2B-dataplane-expert-GGUF",
	filename="gemma-4-E2B-dataplane-expert-Q4_K_M.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

Use Docker

docker model run hf.co/mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mhndayesh/gemma-4-E2B-dataplane-expert-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": "mhndayesh/gemma-4-E2B-dataplane-expert-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

Ollama
How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with Ollama:
```
ollama run hf.co/mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
```

Unsloth Studio

How to use mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for mhndayesh/gemma-4-E2B-dataplane-expert-GGUF to start chatting

How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-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 mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mhndayesh/gemma-4-E2B-dataplane-expert-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 "mhndayesh/gemma-4-E2B-dataplane-expert-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"

Docker Model Runner
How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with Docker Model Runner:
```
docker model run hf.co/mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M
```

Lemonade

How to use mhndayesh/gemma-4-E2B-dataplane-expert-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull mhndayesh/gemma-4-E2B-dataplane-expert-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.gemma-4-E2B-dataplane-expert-GGUF-Q4_K_M

List all available models

lemonade list

⚙️ Recommended runtime settings — gemma-native sampling temperature 1.0, top_k 64, top_p 0.95, min_p 0.01 (the min_p 0.01 floor prevents the reasoning-loop empty-answer issue), context length ≥ 16k (32k recommended), and a generous max_tokens when running with thinking on. The Gemma-4 thinking path needs --jinja.

gemma-4-E2B-dataplane-expert (GGUF)

Base gemma-4-E2B-it with an eBPF & the Networking Data Plane FactBank baked into its chat-template. The model answers correctly about post-cutoff / breaking-change APIs in 7 networking data-plane libraries — not by fine-tuning, but by carrying a searchable bank of 318 landmine facts that fires inside llama.cpp at inference time. Weights are untouched (only the GGUF chat-template was rewritten); no external RAG service.

The model supplies the reasoning; the bank supplies the knowledge it was never trained on.

🔗 Full project — all experts, methodology, per-question transcripts, and benchmarks: github.com/mhndayesh/experts-models

What this expert is about

The Linux networking data plane — eBPF/XDP in the kernel, DPDK/VPP in userspace, and the controllers (Cilium, FRR) that program them — churns hard: the whole BCC→libbpf (CO-RE) shift, libxdp splitting out of the kernel tree, Cilium config keys renamed (tunnel→routingMode). This bank carries those breaking-change facts so the model stops answering from the old world.

Everything here is a landmine fact: post-cutoff, reverses-a-trained-habit, or a silent failure the model wouldn't otherwise catch. Every fact is quote-grounded to a real migration guide / changelog line. Examples the base gets wrong and this model gets right: the BCC→libbpf CO-RE conversion, libxdp's split from the kernel, Cilium's tunnel→routingMode, DPDK removed APIs, FRR operator changes.

Libraries in the bank (7) — 318 facts total

library	facts	what it is / the churn
`libbpf`	93	the canonical eBPF loader — v1 API, BCC→libbpf migration
`cilium`	74	eBPF-based Kubernetes networking/CNI — config-key renames
`frr`	65	FRRouting (BGP/OSPF/… routing suite) — operator-facing changes
`ebpf`	42	eBPF programming model — the BCC→libbpf CO-RE shift
`dpdk`	25	userspace packet processing — deprecated/removed APIs
`xdp`	12	XDP / libxdp fast path — the split from the kernel tree
`vpp`	7	FD.io Vector Packet Processing — release changes

Where the facts come from (mined sources)

Each library's facts were extracted from its migration guide / changelog (source targeting is the whole game — a migration guide, not release-note noise), then quote-verified against the source line:

eBPF BCC→libbpf migration guide
libbpf v1 notes
xdp/libxdp split doc
cilium upgrade guide
DPDK deprecation notices
VPP release notes
FRR changelog

Full provenance (the mined source docs themselves) lives in the repo under v2/extractor/experts/ebpf-dataplane/sources/.

Results (this model — hand-verified)

Same 47 adversarial landmine questions, base vs. this baked model, identical prompts (the bank injects in-engine). Config: thinking-on + authority framing, Gemma-native sampling.

	base E2B	this model	Δ	error-closure*
/47 landmine questions	5/47 (10.6%)	32/47 (68.1%)	+27	64%

* of the answers the base got wrong, the fraction the bank fixed. Every answer was hand-scored — an automated substring check miscounts in both directions (it fails a correct answer that names the old API as a contrast, and passes a semantically wrong one).

Honest note (12B): ~3 of 47 cases regressed — the base was already right and the injected fact + reasoning made it worse (the reasoning-paradox tax). Reported, not hidden.

How to run

The bank lives in the chat-template, so retrieval needs the template applied — run on llama.cpp:

llama-server -m gemma-4-E2B-dataplane-expert-Q4_K_M.gguf --jinja --port 8080 --ctx-size 8192

Then query normally — the same prompt you'd send the base model; the bank fires automatically for covered topics. Sampling — use Gemma-native, not a bare low temperature: temperature 1.0, top_k 64, top_p 0.95, min_p 0.01 (the min_p floor prevents reasoning-loop empty answers).

Best accuracy (authority + thinking) — this is how the numbers above were measured. Send chat_template_kwargs={"enable_thinking": true} and a system prompt telling the model the looked-up facts are verified and supersede its training. A reasoning model otherwise tends to "correct" an injected fact back to its trained prior; authority framing holds the fact.

Limitations

Scoped to the 7 covered libraries. Outside them it's the base model.
Supplies knowledge, not reasoning. A multi-step transform can still fail even with the right fact retrieved.
Retrieval gate is token-based, with aliases. This bake includes the gate-alias fix — a rename's OLD name (e.g. "CrackMapExec") also opens the tab — but a wholly unrelated phrasing may still miss.
Numbers are hand-scored landmine tests, not a general coding benchmark.

Papers

The write-ups behind this project (PDFs, rendered on GitHub): Research Report · Idea · Technical · Verdict · Evidence Ledger

Provenance & license

Base: lmstudio-community/gemma-4-E2B-it-GGUF (Q4_K_M). This model = that GGUF with tokenizer.chat_template rewritten to embed an inverted-index retriever + the bank (factbank.version 0.4.0, gate-alias fix applied).
License: Google Gemma Terms of Use (license: gemma) — a gemma-4 derivative. The fact bank is from the FactBank project (see the repo LICENSE); mined sources keep their own licenses.
Source & method: github.com/mhndayesh/experts-models.