Add Gemma 4 E2B recipes (MobiusModelBuilder + INT4)

[justinchuby]

Olive recipes for google/gemma-4-E2B-it using the MobiusModelBuilder pass (Olive PR microsoft/Olive#2406).

Recipes

Config	Pipeline	Output
gemma4_fp32_cpu.json	MobiusModelBuilder (fp32)	4 ONNX components (~5GB)
gemma4_int4_cuda.json	MobiusModelBuilder (fp16) → OnnxBlockWiseRtnQuantization (int4)	4 quantized ONNX components (~2.8GB)
gemma4_int4_kquant_cpu.json	MobiusModelBuilder (fp32) → OnnxKQuantQuantization (int4)	4 quantized ONNX components (k-quant, CPU)

Prerequisites

pip install olive-ai[gpu] mobius-ai

Validated

INT4 CUDA pipeline tested end-to-end (~3.5 min):

• 97-98% of weights quantized (MatMulNBits + GatherBlockQuantized)
• 4 components: decoder (2.4G), audio (152M), embedding (199M), vision (89M)

[Copilot]

Pull request overview

Adds Olive recipe configs for building and quantizing the HuggingFace google/gemma-4-E2B-it model using the MobiusModelBuilder pass, including an INT4 CUDA pipeline.

Changes:

• Added FP32 CPU build recipe (gemma4_fp32_cpu.json) using MobiusModelBuilder.
• Added FP16 build + INT4 RTN quantization CUDA recipe (gemma4_int4_cuda.json) using OnnxBlockWiseRtnQuantization.
• Added model-specific README documenting prerequisites and how to run both recipes.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
google-gemma-4-E2B-it/gemma4_int4_cuda.json	New CUDA recipe: build FP16 with Mobius, then INT4 block-wise RTN quantization.
google-gemma-4-E2B-it/gemma4_fp32_cpu.json	New CPU recipe: build FP32 with Mobius.
google-gemma-4-E2B-it/README.md	Usage docs for running the two new recipes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[devang-ml]

Please add info.yml and LICENSE file.

[justinchuby]

@copilot Add info.yml and LICENSE file.

[Copilot]

@copilot Add info.yml and LICENSE file.

Added info.yml and LICENSE (Apache 2.0, matching the google-deepmind/gemma repo) to google-gemma-4-E2B-it/ in commit 58b231a.

Please add info.yml and LICENSE file.

Added info.yml and LICENSE (Apache 2.0) to google-gemma-4-E2B-it/ in commit 58b231a.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• ai.google.dev
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node /home/REDACTED/work/_temp/ghcca-node/node/bin/node --enable-source-maps /home/REDACTED/work/_temp/copilot-developer-action-main/dist/index.js (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)