{"slug": "google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on", "title": "Google DeepMind Releases Gemma 4 QAT Checkpoints: Q4_0 and a New Mobile Format Cut On-Device Memory", "summary": "Google DeepMind released Quantization-Aware Training (QAT) checkpoints for the Gemma 4 model family, targeting local deployment on edge devices and consumer GPUs. The new Q4_0 QAT format reduces memory footprint to 3.2 GB for the E2B model and 5 GB for the E4B model, while a new mobile QAT schema compresses the E2B model to approximately 1 GB. The release follows the Gemma 4 launch in April and aims to improve quality preservation compared to standard post-training quantization at the same model sizes.", "body_md": "Google DeepMind released Quantization-Aware Training (QAT) checkpoints for the Gemma 4 family. The release targets local deployment on edge devices and consumer GPUs. It follows the Gemma 4 launch in April and a 12B model two days earlier.\n\nWe compared the available Gemma 4 edge-model formats using only published numbers. The goal was simple. Show what each precision level costs in memory. Then show what QAT actually changes.\n\n**What QAT actually does**\n\nQuantization shrinks a model by lowering weight precision. Standard Post-Training Quantization (PTQ) compresses a finished model. That often degrades quality. QAT instead simulates quantization during training. The model learns to compensate for the precision loss.\n\nGoogle’s AI team states its QAT results yield higher overall quality than standard PTQ baselines. Google did not publish Gemma 4 QAT benchmark scores in the announcement. For context, Gemma 3 QAT cut the Q4_0 perplexity drop by 54% using llama.cpp evaluation. We cite that only as prior-generation precedent.\n\n**The comparison task**\n\nCompare Gemma 4 E2B and E4B across three formats. The formats are BF16, Q4_0 QAT, and the new mobile QAT schema. Rank them on memory footprint, quality preservation, and on-device accessibility. Use published figures only.\n\n**Memory results**\n\n| Format | E2B | E4B | Basis |\n|---|---|---|---|\n| BF16 (16-bit) | 9.6 GB | 15 GB | Official Gemma 4 docs |\n| Q4_0 (4-bit, QAT) | 3.2 GB | 5 GB | Official Gemma 4 docs |\n| Mobile (QAT, E2B) | ~1 GB | — | QAT announcement |\n\nThe Q4_0 figures match the footprint of PTQ Q4_0. QAT does not change the size at a given format. It improves quality at that size. The new mobile schema delivers the additional reduction.\n\nUsing that mobile schema, Google reduced Gemma 4 E2B to about 1GB. Developers can go lower still. The text-only model without Per-Layer Embeddings needs under 1GB, dropping the audio and vision encoders.\n\n**Per-format breakdown**\n\nBF16 is the quality baseline. E2B needs 9.6 GB and E4B needs 15 GB. It is the reference point, not a phone deployment target.\n\nQ4_0 QAT is the general-purpose local format. E2B drops to 3.2 GB and E4B to 5 GB. QAT preserves more quality here than PTQ at the same size. This format fits consumer GPUs. Earlier E2B testing also ran on a Raspberry Pi 5 at INT4.\n\nThe mobile format is the edge-specialized schema. It brings E2B to about 1 GB. It uses static activations, channel-wise quantization, and targeted 2-bit compression.\n\n**How the mobile schema works**\n\nGoogle AI team engineered four techniques for mobile hardware. Static activations pre-calculate scaling during training, reducing on-device work. Channel-wise quantization fits the design of mobile accelerators. Targeted 2-bit quantization compresses only the token-generation layers. Embedding and KV cache optimization shrinks the active memory footprint.\n\nCore reasoning layers stay at higher precision. That protects capability while cutting storage. Developers can also deploy text-only and drop the audio and vision encoders. That trims memory further for use cases that need no multimodality.\n\n**Dimension breakdown**\n\nScores are a qualitative ranking of the formats for on-device use. Memory is the only hard-measured axis. Quality reflects Google’s disclosed design, not measured Gemma 4 numbers. Each score has a one-line basis.\n\n| Dimension | BF16 | Q4_0 QAT | Mobile QAT |\n|---|---|---|---|\n| Memory footprint | 1 — heaviest, 9.6 GB E2B | 4 — 3.2 GB E2B | 5 — ~1 GB E2B text-only |\n| Quality preservation | 5 — full-precision baseline | 4 — QAT-preserved, near baseline | 3 — 2-bit token layers, core kept higher |\n| Decode speed | 2 — no quantization speedup | 4 — 4-bit accelerates decode | 5 — mobile-optimized static activations |\n| Deployment breadth | 4 — loadable but heavy | 5 — llama.cpp, Ollama, LM Studio, vLLM, MLX | 3 — LiteRT-LM, Transformers.js, edge-focused |\n| On-device accessibility | 1 — needs large GPU | 4 — consumer GPU, Raspberry Pi 5 | 5 — runs on phones |\nTotal (/25) | 13 | 21 | 21 |\n\n**Winner**\n\nThe result is a tie by design. Q4_0 QAT and mobile QAT both score 21, but for different hardware. For phones, the mobile format leads. It reaches about 1GB on E2B and targets mobile accelerators directly. For laptops and consumer GPUs, Q4_0 QAT is the practical default. BF16 stays the quality reference, not a local choice.\n\n**Methodology and limits**\n\nMemory figures come from Google’s Gemma 4 documentation. The ~1GB E2B figure comes from the QAT announcement. Quality is Google’s stated claim. No independent Gemma 4 QAT quality numbers were published at release. We did not run the models locally for this comparison. Developers should test at their own quantization and workload before building.\n\n**Key Takeaways**\n\n- Q4_0 QAT cuts Gemma 4 E2B to 3.2 GB and E4B to 5 GB, from 9.6 GB and 15 GB at BF16.\n- A new mobile QAT schema brings E2B to about 1 GB; text-only without PLE goes under 1 GB.\n- QAT changes quality at a given size, not the size itself; the mobile format drives the extra memory cut.\n- Google claims higher quality than PTQ but published no Gemma 4 QAT benchmark numbers at release.\n- Weights ship today on Hugging Face with llama.cpp, Ollama, LM Studio, vLLM, MLX, and LiteRT-LM support.\n\n**Marktechpost’s Visual Explainer**\n\nCheck out the **Model weights **(** Q4_0 QAT collection, Mobile QAT collection)** and\n\n[Google blog (QAT release)](https://blog.google/innovation-and-ai/technology/developers-tools/quantization-aware-training-gemma-4/)\n\n**.** Also, feel free to follow us on\n\n**and don’t forget to join our**[Twitter](https://x.com/intent/follow?screen_name=marktechpost)\n\n**and Subscribe to**\n\n[150k+ ML SubReddit](https://www.reddit.com/r/machinelearningnews/)**. Wait! are you on telegram?**\n\n[our Newsletter](https://www.aidevsignals.com/)\n\n[now you can join us on telegram as well.](https://t.me/machinelearningresearchnews)Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? [Connect with us](https://forms.gle/wbash1wF6efRj8G58)", "url": "https://wpnews.pro/news/google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on", "canonical_source": "https://www.marktechpost.com/2026/06/05/google-deepmind-releases-gemma-4-qat-checkpoints-q4_0-and-a-new-mobile-format-cut-on-device-memory/", "published_at": "2026-06-05 18:59:38+00:00", "updated_at": "2026-06-05 19:03:39.938869+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-research", "ai-products"], "entities": ["Google DeepMind", "Gemma 4", "Gemma 3", "QAT", "llama.cpp"], "alternates": {"html": "https://wpnews.pro/news/google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on", "markdown": "https://wpnews.pro/news/google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on.md", "text": "https://wpnews.pro/news/google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on.txt", "jsonld": "https://wpnews.pro/news/google-deepmind-releases-gemma-4-qat-checkpoints-q4-0-and-a-new-mobile-format-on.jsonld"}}