Brieftide
Topic hub
Model Compression
Covers methods to shrink and speed AI models, including pruning, quantization, distillation, MoE compression, and train and inference alignment.
77 briefs
Model Compression · Page 3
- Serverless GPUs add Opus 4.7 Fast and Qwen Image 2.0 supportThe Brieftide
- OpenAI Codex with GPT-5.5 used to ship production systemsThe Brieftide
- Parameter Golf challenge draws 1,000+ participants, 2,000+The Brieftide
- MIT WRING: debiasing vision models without bias amplificationThe Brieftide
- Privacy-preserving on-device AI: MIT method enables trainingThe Brieftide
- EnergAIzer by MIT: fast AI power estimates in secondsThe Brieftide
- Transformers.js for Chrome extensions: how to run modelsThe Brieftide
- Hugging Face launches transformers-to-mlx converter for modelsThe Brieftide
- MIT pruning method trims models during training, reduces computeThe Brieftide
- mRNA language models: OpenMed trains 25-species models for $165The Brieftide
- OpenClaw: Hugging Face releases open-source LLM fork v1.0The Brieftide
- Hugging Face and NVIDIA: build domain embeddings in a dayThe Brieftide
- Ulysses Sequence Parallelism: Million-Token TrainingThe Brieftide
- DeepMind Nano Banana 2 release: image model at Flash speedThe Brieftide
- Mixture of Experts (MoE) in Transformers: Hugging Face guideThe Brieftide
- Information-driven imaging: Berkeley releases noisy-data estimatorThe Brieftide
- AlphaFold DeepMind: engineering heat-tolerant Rubisco for cropsThe Brieftide
- DeepSeek V3 to V3.2: architecture, sparse attention, RL updatesThe Brieftide
- Nano Banana Pro: DeepMind's Gemini 3 Pro Image model launchThe Brieftide
- DeepSeek V3: 671B model, MLA and MoE architectural choices, 2025The Brieftide
- PEVA whole-body egocentric video prediction with 16s rolloutsThe Brieftide
- SecAlign and StruQ: Berkeley AI defenses cut prompt-injectionThe Brieftide
