# Google DeepMind 发布 Gemma 4 QAT 检查点：Q4_0 和新的移动格式降低设备端内存 - 来源：MarkTechPost（RSS） - 作者：Asif Razzaq - 发布时间：2026-06-06 02:59 - AIHOT 分数：50 - AIHOT 链接：https://aihot.virxact.com/items/cmq1bgq8z0e9osltrqyy4digz - 原文链接：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 ## AI 摘要 Google DeepMind 推出 Gemma 4 量化感知训练（QAT）检查点，包含 Q4_0 格式和一种新的移动端格式，旨在降低设备端内存占用。对比 BF16、Q4_0 QAT 和移动版 QAT 三种边缘端格式，官方公布了各格式的内存数据与设计权衡。 ## 正文 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. We 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. What QAT actually does Quantization 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. Google’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. The comparison task Compare 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. Memory results FormatE2BE4BBasis BF16 (16-bit)9.6 GB15 GBOfficial Gemma 4 docs Q4_0 (4-bit, QAT)3.2 GB5 GBOfficial Gemma 4 docs Mobile (QAT, E2B)~1 GB—QAT announcement The 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. Using 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. Per-format breakdown BF16 is the quality baseline. E2B needs 9.6 GB and E4B needs 15 GB. It is the reference point, not a phone deployment target. Q4_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. The 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. How the mobile schema works Google 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. Core 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. Dimension breakdown Scores 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. DimensionBF16Q4_0 QATMobile QAT Memory footprint1 — heaviest, 9.6 GB E2B4 — 3.2 GB E2B5 — ~1 GB E2B text-only Quality preservation5 — full-precision baseline4 — QAT-preserved, near baseline3 — 2-bit token layers, core kept higher Decode speed2 — no quantization speedup4 — 4-bit accelerates decode5 — mobile-optimized static activations Deployment breadth4 — loadable but heavy5 — llama.cpp, Ollama, LM Studio, vLLM, MLX3 — LiteRT-LM, Transformers.js, edge-focused On-device accessibility1 — needs large GPU4 — consumer GPU, Raspberry Pi 55 — runs on phones Total (/25)132121 Winner The 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. Methodology and limits Memory 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. Key Takeaways 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. A new mobile QAT schema brings E2B to about 1 GB; text-only without PLE goes under 1 GB. QAT changes quality at a given size, not the size itself; the mobile format drives the extra memory cut. Google claims higher quality than PTQ but published no Gemma 4 QAT benchmark numbers at release. Weights ship today on Hugging Face with llama.cpp, Ollama, LM Studio, vLLM, MLX, and LiteRT-LM support. Marktechpost’s Visual Explainer Marktechpost · Benchmark Gemma 4 QAT: Comparing Q4_0 and the New Mobile Format Google DeepMind released Quantization-Aware Training checkpoints for Gemma 4. We compared three edge-model formats on published numbers. Formats compared BF16 (16-bit) · Q4_0 QAT (4-bit) · Mobile QAT June 5, 2026 The Comparison Task What we ranked $ compare gemma-4 --models E2B,E4B \ --formats BF16,Q4_0-QAT,MOBILE-QAT \ --rank memory,quality,accessibility \ --source published-only --no-self-run Memory from official Gemma 4 docs. Quality from Google’s stated claim. No models run locally. Format 1 of 3 · Reference BF16 (16-bit) 13 / 25 The full-precision quality baseline. E2B needs 9.6 GB and E4B needs 15 GB. Top observation: a reference point, not a phone or laptop deployment target. Format 2 of 3 · Laptop / GPU Q4_0 QAT (4-bit) 21 / 25 The general-purpose local format. E2B drops to 3.2 GB and E4B to 5 GB. Top observation: QAT preserves more quality than PTQ at the same 4-bit size. Format 3 of 3 · Mobile Mobile QAT 21 / 25 The edge-specialized schema. Brings E2B to about 1 GB. Top observation: 2-bit on token layers, reasoning layers kept at higher precision. Leaderboard Full ranking DimensionBF16Q4_0 QATMobile QAT Memory footprint145 Quality preservation543 Decode speed245 Deployment breadth453 On-device accessibility145 Total132121 Tie by design: Q4_0 wins laptops and GPUs; mobile wins phones. Key Takeaways What developers should know Q4_0 QAT cuts E2B to 3.2 GB and E4B to 5 GB, from 9.6 GB and 15 GB at BF16. A new mobile QAT schema brings E2B to about 1 GB; text-only without PLE goes under 1 GB. QAT changes quality at a given size; the mobile format drives the extra memory cut. Google claims higher quality than PTQ but published no Gemma 4 QAT numbers. Weights ship today on Hugging Face with llama.cpp, Ollama, vLLM, and MLX support. ‹ Prev Next › Memory: official Gemma 4 documentation. ~1 GB E2B: QAT announcement (mobile format); text-only without PLE is under 1 GB. Quality: Google’s stated claim — no independent Gemma 4 QAT scores at release. Check out the Model weights (Q4_0 QAT collection, Mobile QAT collection) and Google blog (QAT release). Also, feel free to follow us on Twitter and don’t forget to join our 150k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Google DeepMind Releases Gemma 4 QAT Checkpoints: Q4_0 and a New Mobile Format Cut On-Device Memory appeared first on MarkTechPost.