corpus.blog
41554 blogs · [ { "id": "01a08cbd-11e4-7072-b051-e8248c0b0dea", "title": "How Not to Build Another Megatron (Part 2): A Minimalist Parallelism Scheme", "url": "https://le.qun.ch/en/blog/2026/09/08/minimalism-parallelism/", "published_at": "2026-09-08T07:00:00+00:00" }, { "id": "01a08cbd-11e4-7072-b051-e8248cf9af6c", "title": "How Not to Build Another Megatron (Part 1): How Pipeline Parallelism Contaminates the Entire Training Framework", "url": "https://le.qun.ch/en/blog/2026/09/07/no-pp/", "published_at": "2026-09-07T07:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a95fe8d34c", "title": "Explorations of RDMA in LLM Systems", "url": "https://le.qun.ch/en/blog/2025/11/09/rdma-p2p-for-llm/", "published_at": "2025-11-09T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a960110772", "title": "Quick Follow-up on Inter-node RL Weight Transfer", "url": "https://le.qun.ch/en/blog/2025/09/17/rl-weight-transfer-2/", "published_at": "2025-09-17T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9605e8856", "title": "Journey to 2-second Inter-node RL Weight Transfer", "url": "https://le.qun.ch/en/blog/2025/09/07/rl-weight-transfer/", "published_at": "2025-09-07T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a961316554", "title": "Harnessing 3200Gbps Network (15): Lazy Posting", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-15-lazy/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9619cb2aa", "title": "Harnessing 3200Gbps Network (14): Batch Posting", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-14-batch/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a96271d354", "title": "Harnessing 3200Gbps Network (13): State Sharding", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-13-shard/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9631f023a", "title": "Harnessing 3200Gbps Network (12): CPU Core Pinning", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-12-pin/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9635cbcad", "title": "Harnessing 3200Gbps Network (11): Multi-threading", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-11-multithread/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a963af130e", "title": "Harnessing 3200Gbps Network (10): Pre-benchmark Warmup", "url": "https://le.qun.ch/en/blog/2024/12/31/libfabric-efa-10-warmup/", "published_at": "2024-12-31T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a963e82508", "title": "Harnessing 3200Gbps Network (9): Using 32 Network Cards", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-9-multinet/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a964618d29", "title": "Harnessing 3200Gbps Network (8): Bus Topology", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-8-topo/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a965344c19", "title": "Harnessing 3200Gbps Network (7): Queuing and Benchmark", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-7-queue/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a965a22ec6", "title": "Harnessing 3200Gbps Network (6): GPUDirect RDMA WRITE", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-6-write/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a965ba9557", "title": "Harnessing 3200Gbps Network (5): Bidirectional SEND and RECV", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-5-bidi/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a96630498a", "title": "Harnessing 3200Gbps Network (4): Unidirectional SEND and RECV", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-4-recv-send/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a966a406fd", "title": "Harnessing 3200Gbps Network (3): libfabric", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-3-fi_info/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a96761718e", "title": "Harnessing 3200Gbps Network (2): High-Performance Network System Design Philosophy", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-2-philosophy/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9681a01cf", "title": "Harnessing 3200 Gbps Network (1): RDMA and EFA", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-1-rdma/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a968cc91db", "title": "Harnessing 3200 Gbps Network: A Journey with RDMA, EFA, and libfabric", "url": "https://le.qun.ch/en/blog/2024/12/25/libfabric-efa-0-intro/", "published_at": "2024-12-25T08:00:00+00:00" }, { "id": "01a0879a-04c3-735b-9cb5-83a9694842fe", "title": "Potentials of Multitenancy Fine-Tuned LLM Serving", "url": "https://le.qun.ch/en/blog/2023/09/11/multi-lora-potentials/", "published_at": "2023-09-11T08:00:00+00:00" } ] posts Claim your blog
Back to le.qun.ch
Blog · corpus.blog/blogs/le.qun.ch/posts

le.qun.ch

le.qun.ch

2026

How Not to Build Another Megatron (Part 2): A Minimalist Parallelism Schemeoriginal ↗
8 Sept 2026
How Not to Build Another Megatron (Part 1): How Pipeline Parallelism Contaminates the Entire Training Frameworkoriginal ↗
7 Sept 2026

2025

Explorations of RDMA in LLM Systemsoriginal ↗
9 Nov 2025
Quick Follow-up on Inter-node RL Weight Transferoriginal ↗
17 Sept 2025
Journey to 2-second Inter-node RL Weight Transferoriginal ↗
7 Sept 2025

2024

Harnessing 3200Gbps Network (15): Lazy Postingoriginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (14): Batch Postingoriginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (13): State Shardingoriginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (12): CPU Core Pinningoriginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (11): Multi-threadingoriginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (10): Pre-benchmark Warmuporiginal ↗
31 Dec 2024
Harnessing 3200Gbps Network (9): Using 32 Network Cardsoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (8): Bus Topologyoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (7): Queuing and Benchmarkoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (6): GPUDirect RDMA WRITEoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (5): Bidirectional SEND and RECVoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (4): Unidirectional SEND and RECVoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (3): libfabricoriginal ↗
25 Dec 2024
Harnessing 3200Gbps Network (2): High-Performance Network System Design Philosophyoriginal ↗
25 Dec 2024
Harnessing 3200 Gbps Network (1): RDMA and EFAoriginal ↗
25 Dec 2024
Harnessing 3200 Gbps Network: A Journey with RDMA, EFA, and libfabricoriginal ↗
25 Dec 2024

2023

Potentials of Multitenancy Fine-Tuned LLM Servingoriginal ↗
11 Sept 2023