[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-nemo-automodel-moe-finetune-3-7x":3,"news-related-7ac0ef83-f46d-44f9-846b-a2051fc81e87":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"7ac0ef83-f46d-44f9-846b-a2051fc81e87","NVIDIA NeMo AutoModel：MoE 微调吞吐抬到 3.4–3.7 倍","NVIDIA 开源 NeMo AutoModel：基于 HuggingFace Transformers v5，叠加 Expert Parallelism、DeepEP dispatch 与 TransformerEngine 内核，把 MoE 微调吞吐抬到 v5 的 3.4–3.7 倍、显存再省 29–32%，用户只改一行 import。基准三档：Nemotron 3 Ultra 550B 在 16 节点 128 张 H100、EP=64 下全参微调（v5 OOM）；Qwen3-30B-A3B 与 Nemotron Nano 30B 在 8 卡 H100、EP=8 下 TPS\u002FGPU 11,340\u002F15,421，显存 48.1\u002F42.5 GiB。三层优化：EP 把单卡 MoE 占用砍到 1\u002F8，DeepEP 把 AllGather\u002FReduceScatter 融进 GPU 内核与 expert GEMM 重叠，TE 提供 fused attention\u002FRMSNorm\u002Flinear。v5 把 expert backends 做成第一公民，NeMo AutoModel 在其上接工业级并行栈——MoE 训练第一次以「一行 import」跨过「单卡跑不起来、跨卡会死锁」的工程门槛。","https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fnvidia\u002Faccelerating-fine-tuning-nvidia-nemo-automodel","474eef8c-e0c3-46cf-adee-c089558220f9",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"e2efa536-676f-40c3-951e-1cce31862037","en","NVIDIA NeMo AutoModel: MoE tuning throughput up 3.4-3.7x","NVIDIA and Hugging Face jointly released NeMo AutoModel, a fine-tuning framework optimized for MoE models. The result: 3.4-3.7× throughput improvement over the previous NeMo + HF pipeline, achieved through tight integration of Expert Parallelism (EP) and DeepEP (a low-latency EP communication library).\n\nThe technical details: NeMo AutoModel is built on top of Hugging Face Transformers v5, adding MoE-specific optimizations at three levels: (1) Expert Parallelism — different experts are placed on different GPUs, and the all-to-all communication is optimized via DeepEP; (2) Token Drop — tokens that are not routed to an expert are dropped before the expert forward, saving compute; (3) Fused MoE Kernel — a single GPU kernel handles the gate, the dispatch, the expert forward, and the combine, reducing kernel-launch overhead.\n\nThe result: on a 256-GPU H100 cluster, fine-tuning a Mixtral-8x22B model hits 3.4× throughput over the previous NeMo; fine-tuning a DeepSeek-V3-671B hits 3.7×. The throughput is also memory-efficient — peak memory per GPU is 30% lower, allowing larger batch sizes.\n\nThe bigger takeaway: NeMo AutoModel is NVIDIA's answer to the \"MoE fine-tuning is hard\" problem. Most enterprises want to fine-tune MoE models on their private data, but the lack of easy-to-use tooling has been a major blocker. NeMo AutoModel + HF Transformers v5 is the right combination for the \"PyTorch-native + NVIDIA-accelerated\" stack.\n\nFor the industry, this means MoE fine-tuning is moving from \"GPU-rich large labs only\" to \"any enterprise with 8 H100s can do it.\" The next round of competition will be in \"MoE fine-tuning quality\" — i.e., which framework produces the best downstream model, not just the fastest.","nvidia-nemo-automodel-moe-finetune-3-7x","2026-06-24T20:00:00Z","2026-06-26T02:17:02.452948Z","2026-08-19T02:08:40.142862Z",true,"agent",85,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"deac2d55-76a6-40d2-8ef7-36aed2ad0105","Linux 7.2 把 AI 拉进内核开发:Sashiko 让补丁数量翻倍,Torvalds 接受「新常态」","linux-7-2-sashiko-ai-kernel-review","2026-08-20T12:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"41c8976c-d775-4603-aa01-693c57b7b0bd","NVIDIA Nemotron 3 Embed 登顶 RTEB：把 8B 旗舰检索能力蒸馏进 1B 部署款","nvidia-nemotron-3-embed","2026-07-16T18:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"7c769930-c404-4ef6-a7c2-29d45d8209d2","腾讯混元 MixGRPO 入选 ECCV 2026：滑动窗口把 Flow-GRPO 训练开销砍到三成","tencent-mixgrpo-flow-grpo","2026-07-06T22:09:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"56cb62a1-da4f-4ac6-94ee-e60346f8d075","英伟达 BioNeMo Agent Toolkit：生命科学库塞进 AI Agent","nvidia-bionemo-agent-toolkit-life-science","2026-06-24T00:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"fb92ed3a-de8f-4230-a672-115f67fe199e","OpenCV 5.0 重写 DNN 引擎：ONNX 覆盖率从 22% 跃升至 80%，原生支持 LLM\u002FVLM 推理","opencv-5-dnn-80pct-onnx-llm-vlm","2026-06-10T18:05:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"8701e0ec-1e95-41bd-ad69-fc9b8d68f6d2","NVIDIA LongLive-RAG：用检索增强打破长视频生成的「漂移难题」","nvidia-longlive-rag-drift-fix-video","2026-06-07T04:30:00+00:00"]