Instructions to use Laborator/microlens-gemma4-e2b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Laborator/microlens-gemma4-e2b with Transformers:

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

pipe = pipeline("image-text-to-text", model="Laborator/microlens-gemma4-e2b")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Laborator/microlens-gemma4-e2b", dtype="auto")

llama-cpp-python

How to use Laborator/microlens-gemma4-e2b with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Laborator/microlens-gemma4-e2b",
	filename="gguf/gemma-4-e2b-it.BF16-mmproj.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": [
				{
					"type": "text",
					"text": "Describe this image in one sentence."
				},
				{
					"type": "image_url",
					"image_url": {
						"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
					}
				}
			]
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Laborator/microlens-gemma4-e2b with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Laborator/microlens-gemma4-e2b:BF16
# Run inference directly in the terminal:
llama-cli -hf Laborator/microlens-gemma4-e2b:BF16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Laborator/microlens-gemma4-e2b:BF16
# Run inference directly in the terminal:
llama-cli -hf Laborator/microlens-gemma4-e2b:BF16

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 Laborator/microlens-gemma4-e2b:BF16
# Run inference directly in the terminal:
./llama-cli -hf Laborator/microlens-gemma4-e2b:BF16

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 Laborator/microlens-gemma4-e2b:BF16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Laborator/microlens-gemma4-e2b:BF16

Use Docker

docker model run hf.co/Laborator/microlens-gemma4-e2b:BF16

LM Studio
Jan

vLLM

How to use Laborator/microlens-gemma4-e2b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Laborator/microlens-gemma4-e2b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Laborator/microlens-gemma4-e2b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Laborator/microlens-gemma4-e2b:BF16

SGLang

How to use Laborator/microlens-gemma4-e2b 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 "Laborator/microlens-gemma4-e2b" \
    --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": "Laborator/microlens-gemma4-e2b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "Laborator/microlens-gemma4-e2b" \
        --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": "Laborator/microlens-gemma4-e2b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Ollama
How to use Laborator/microlens-gemma4-e2b with Ollama:
```
ollama run hf.co/Laborator/microlens-gemma4-e2b:BF16
```

Unsloth Studio new

How to use Laborator/microlens-gemma4-e2b 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 Laborator/microlens-gemma4-e2b 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 Laborator/microlens-gemma4-e2b to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Laborator/microlens-gemma4-e2b to start chatting

Pi new

How to use Laborator/microlens-gemma4-e2b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Laborator/microlens-gemma4-e2b:BF16

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": "microlens-gemma4-e2b"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Laborator/microlens-gemma4-e2b with Docker Model Runner:
```
docker model run hf.co/Laborator/microlens-gemma4-e2b:BF16
```

Lemonade

How to use Laborator/microlens-gemma4-e2b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Laborator/microlens-gemma4-e2b:BF16

Run and chat with the model

lemonade run user.microlens-gemma4-e2b-BF16

List all available models

lemonade list

MicroLens · Microscopy Vision-Language Model

Gemma 4 E2B fine-tuned on 122,399 microscopy VQA pairs across 145+ genera — diatoms, freshwater and marine zooplankton, fungal spores, fish larvae. Three open versions, all Apache 2.0, all running 100 % offline.

Base model: unsloth/gemma-4-E2B-it (4.65 B params)
Source code: SergheiBrinza/microlens
Android APK: GitHub Releases
Browser demo: HuggingFace Space
Ollama Hub: brinzaengineeringai/microlens-v{1,2,3}
Submission for: Kaggle Gemma 4 Good Hackathon 2026 — Health & Sciences track
License: Apache 2.0 (code, weights, datasets — see LICENSE_AUDIT.md)

Model versions in this repository

Version	Style	When to use
v1	Naive baseline	Historical reference, training-progression demo
v2	Production brief — single line	Pipelines, automated databases, low-latency
v3	Production rich — four sections (genus + morphology + habitat + ID cues)	Citizen science, classroom, judge-facing UIs

Each version ships as GGUF Q4_K_M (~~3.4 GB) + BF16 mmproj (~~940 MB) for llama.cpp / Ollama / mtmd:

gguf/
├── gemma-4-e2b-it.Q4_K_M.gguf       # v1 model (3.4 GB)
├── gemma-4-e2b-it.BF16-mmproj.gguf  # v1 mmproj (940 MB)
├── v2/
│   ├── model.gguf                   # v2 model (3.4 GB)
│   └── mmproj.gguf                  # v2 mmproj (940 MB)
└── v3/
    ├── model.gguf                   # v3 model (3.4 GB)
    └── mmproj.gguf                  # v3 mmproj (940 MB)

Plus the LoRA adapter, merged FP16 model, tokenizer, and processor for downstream training.

Run it

# easiest — Ollama
ollama run brinzaengineeringai/microlens-v3

# stock llama.cpp
llama-server \
    -m gguf/v3/model.gguf \
    --mmproj gguf/v3/mmproj.gguf \
    -ngl 99

Training summary

Metric	v2	v3
Base model	unsloth/gemma-4-E2B-it (4.65 B)	resumed from v2 checkpoint-18351
Trainable params	LoRA r=32, α=64	same adapter, 1 extra epoch
Train samples	99,215	99,215 (rich-format augmented from KB)
Validation samples (held-out)	12,331	220 stratified
Epochs	3	1
Hardware	RTX 3090 Ti	RTX 3090 Ti
Training time	37 h	10 h 16 m
Final eval loss	≈ 0.21	0.0213
Framework	Unsloth FastVisionModel · 4-bit QLoRA	same

Evaluation

Stratified evaluation on held-out val.jsonl (n = 220 across 5 categories):

Metric	v2	v3
Category accuracy	97 – 100 %	97 – 100 %
Genus accuracy (145+ genera, random ≈ 0.7 %)	~ 45 %	~ 45 % (no drift)
Rich-format compliance	n/a	100 %

Datasets

122,399 image–question–answer triples from public, licence-clean microscopy collections:

Category	Source	Samples
Diatoms	UDE Diatoms in the Wild 2024, DiAtlas	97 k
Freshwater zooplankton	ZooLake (EAWAG)	12 k
Marine zooplankton	V1-Zooplankton	10 k
Fungal spores	TGFC fungal	5 k
Fish larvae	ZooLake	0.2 k
Negative-class (no specimen)	synthetic	1.5 k

145+ genera with long-tail handled by category-level fallback. Top-30 genera have hand-curated knowledge-base entries (morphology + habitat + identification cues from AlgaeBase, WoRMS, ITIS, Round 1990, Krammer-Lange-Bertalot 1986–1991). Full per-source attribution: docs/LICENSE_AUDIT.md. No verbatim copyrighted text was used; KB phrasing is original paraphrase.

Limitations

Long-tail genera (~100 of 145+ have <100 training samples each) get category-generic morphology rather than genus-specific descriptions. The 30 most-common genera have hand-curated entries.
Pseudo-class Fish has no species-level annotation in the training data; v3 falls back to a category-level templated description.
Mobile cold-start on sub-$100 phones takes ~3-4 minutes (model load + 292-token prefill on 6 CPU cores). Subsequent inferences in the same session are much faster. Use desktop or browser for real-time work.
English-first scientific output. Translation through the vanilla Gemma 4 base covers 140+ languages but may simplify rare technical terms.

⚠️ Legal — research and educational artefact

Research model · Apache 2.0 · Not a medical device · Not a certified instrument · Use at your own risk.

This model is NOT a medical device, in-vitro diagnostic (IVD), clinical decision-support tool, or regulatory-compliant water-quality measurement instrument (no ISO 17025, EPA, EU Water Framework Directive, or equivalent certification). It is not a substitute for a trained taxonomist or accredited laboratory analysis, and not a calibrated, validated, or peer-reviewed analytical method.

The model's output is a statistical pattern match against the training data distribution, rendered through learned scientific phrasing — not a physical or analytical measurement, not a peer-reviewed identification. The model can be confidently wrong, particularly on:

specimens not represented in training (145+ genera ≠ all microscopic life);
damaged, atypical, or out-of-focus images;
subjects from kingdoms or phyla outside the training categories.

No warranty. This software is provided AS IS, without warranty of any kind, express or implied. In no event shall the author or contributors be liable for any claim, damages, or other liability arising from, out of, or in connection with the model. You assume all risk when downloading, deploying, modifying, or using this model. Always have qualified personnel verify any result that informs a regulatory, environmental, clinical, or health-related decision.

EU AI Act positioning (Regulation EU 2024/1689): Per Article 50 you are interacting with an AI system; outputs are AI-generated content. Per Article 2(6) the artefact is published under the research exemption. It is not a high-risk AI system within the meaning of Annex III, not a safety component of an Annex I product, and does not engage in any practice prohibited by Article 5. Gemma 4 (the upstream GPAI model) is provided by Google DeepMind; obligations under Chapter V rest with the GPAI provider, not the downstream researcher.

Full legal documents: