Unverified50% confidenceEventExact time
Google DeepMind released Gemma 4 Quantization-Aware Training (QAT) checkpoints along with a new mobile quantization format
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
Related Claims
UnverifiedGoogle DeepMind发布Gemma 4量化感知训练（QAT）CheckPoint，通过定向二比特压缩等技术使Gemma 4 E-to-B模型内存占用压缩至约1GB级别，相关权重已上线Hugging Face68% similarUnverifiedGoogle DeepMind发布Gemma 4量化感知训练权重，通过定向2Bit压缩技术将12B参数模型的内存占用降至约1GB65% similarUnverifiedUnsloth核心技术包括智能梯度检查点、4-bit量化感知训练QLoRA的深度优化，以及针对Llama、Mistral、Gemma等架构的专项适配64% similarVerifiedGoogle DeepMind在开发Gemini时采用了从头开始的多模态训练策略，使其能够直接理解视频中的时序变化和音频中的音调情感等非文本信息64% similarVerifiedDeepSeek由量化私募巨头幻方科技孵化，其R1系列模型以极低训练成本实现接近GPT-4级别的推理能力63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/53679API
curl https://kongchang.com/api/v1/knowledge/claims/53679MCP
get_claim(id=53679)