Gemma 4 E2B — Cactus engine bundle

Gemma 4 E2B instruction-tuned (vision and audio included), packaged for the Cactus on-device engine.

A small, on-device model is fast and private, but sometimes wrong. At Cactus we post-train models to know when they are wrong: we ship probes inside the checkpoint that score every answer with a confidence between 0 and 1, returned as structured data (never parsed out of the answer text). Answer on-device when confidence is high; re-route to a bigger model when it's low:

if confidence < 0.85:
    answer = ask_a_bigger_model(prompt)

Benchmarks

Gemma 4 E2B Hybrid, the smallest Gemma model, matches Gemini 3.1 Flash-Lite on most benchmarks by routing only 15–35% of queries to Flash-Lite and running the rest itself:

Benchmark	Handoff to match Flash-Lite (FP16)	At 4-bit	At 3-bit
ChartQA	15–20%	25–30%	40–50%
MMBench	30–35%	40–45%	50–55%
LibriSpeech	25–30%	35–40%	55–65%
GigaSpeech	30–35%	40–45%	50–55%
MMAU	30–35%	35–40%	50–55%
MMLU-Pro	45–55%	~90%	n/a

Quantisation quality is measured on Cactus Quants, which performs well at uniform quantization; developers are encouraged to benchmark Unsloth, GGUF, and MLX quantization independently.

Quickstart

# pip install cactus-compute
import json
from cactus.bindings.cactus import cactus_complete, cactus_init
from cactus.cli.download import download_bundle

lm = cactus_init(str(download_bundle("Cactus-Compute/gemma-4-E2B-it")))
result = cactus_complete(
    lm,
    [{"role": "user", "content": "What is the capital of France?"}],
    json.dumps({"max_tokens": 512, "auto_handoff": False}),
    None,
    lambda *_: None,
)
print(result["response"].strip())
print("confidence:", result["confidence"])

"auto_handoff": false returns the raw probe score so your app decides when to escalate; leave it on to let the engine hand off by itself.

Routing quality (AUROC)

AUROC measures how well the probe separates wrong answers from right ones (higher = better, 0.5 is random, 1.0 is perfect):

Hold-out	Modality	Cactus Hybrid	Token Entropy
MMLU	text MCQ	0.770	0.697
MMLU-Pro	text MCQ	0.771	0.692
ARC-Easy	text MCQ	0.888	0.655
ARC-Challenge	text MCQ	0.834	0.646
GSM8K (3-shot)	text gen	0.782	0.731
MMBench-EN-Dev	vision MCQ	0.840	0.435
ChartQA	vision QA	0.779	0.615
DocVQA	vision QA	0.781	0.512
MMAU	audio MCQ	0.789	0.517
GigaSpeech	audio	0.876	0.343
Earnings-22	audio	0.839	0.323
LibriSpeech	audio	0.822	0.427
Mean		0.814	0.549

The strongest result: the probe was trained on zero audio data, yet achieves 0.79–0.88 AUROC on four audio benchmarks (two transcription, one audio MCQ, one out-of-domain transcription). This rules out surface-level explanations: the probe is reading a modality-independent correctness signal from the hidden state, not memorizing patterns from training data.