Gemma 4 E4B Resilient Vision
This model is a compressed version of Gemma 4 optimized for the Resilient AI Challenge 2026 Image-to-Text category.
Optimization & Inference
- Inference Engine: vLLM (v0.20.2)
- Compression: Native vLLM default optimization and memory management.
- Precision: 4-bit (via vLLM handshake).
Evaluation Setup
The model is served via vLLM with challenge-mandated parameters:
- Temperature: 1.0
- Top-P: 0.95
- Top-K: 64
Evaluation
# python bench_vision.py
================================================================================
GEMMA 4 E4B — vLLM HANDSHAKE EVALUATION
================================================================================
🔐 Determinism Locked | Seed: 123
📦 Handshaking with vLLM: http://127.0.0.1:8000/v1/chat/completions
✓ TOTAL SYSTEM RAM: 3.33 GB
============================================================
📸 [1/3] Bee on Flower
============================================================
📝 Generated: Based on the image provided, here is a precise description:
**Image Description:**
This is a close-up, macro photograph...
⏱️ RTF: 0.0800 s/w | Words: 115 | Throughput: 12.5 w/s
🔋 Energy: 450.71 J | Power: 49.0W
💻 RAM: 3.46 GB | Score: 1.000
============================================================
📸 [2/3] Wisconsin Boardwalk
============================================================
📝 Generated: Based on the image provided, there are **no identifiable individuals** present.
**Precise Image Description:**
This is...
⏱️ RTF: 0.0536 s/w | Words: 117 | Throughput: 18.7 w/s
🔋 Energy: 429.54 J | Power: 68.5W
💻 RAM: 3.46 GB | Score: 1.000
============================================================
📸 [3/3] Turing Award Winners
============================================================
📝 Generated: This image is a composite featuring three prominent figures in the field of Artificial Intelligence (AI) and Machine Lea...
⏱️ RTF: 0.0585 s/w | Words: 106 | Throughput: 17.1 w/s
🔋 Energy: 436.13 J | Power: 70.3W
💻 RAM: 3.46 GB | Score: 1.000
================================================================================
📊 FINAL CHALLENGE AUDIT — GEMMA 4 E4B (vLLM)
================================================================================
Average RAM: 3.46 GB
Average RTF: 0.0640 s/w
Avg Throughput: 16.1 words/sec
Total Energy: 1316.38 J
Total CO2e: 0.000365 kg
Avg Quality: 1.000
🔍 CHALLENGE TARGETS:
RAM < 4GB: ✅ PASS (3.46 GB)
RTF < 1.0: ✅ PASS (0.0640)
Quality > 80%:✅ PASS (1.000)
🎉 AUDIT COMPLETE — ALL TARGETS ACHIEVED!
License
Original Google Gemma License.