Deep Dive Series on Training LLMs from Scratch
Автор: C-DAC
Загружено: 2025-08-04
Просмотров: 983
We are happy to share the recording of the first session from the webinar series jointly organized by NVIDIA and C-DAC, Pune, focused on training large language models (LLMs) from scratch
This multi-part webinar series provides a step-by-step walkthrough of the complete process involved in training Large Language Models (LLMs).
1) Cluster Health Check using NCCL and MLPerf Benchmarks
2) Large-Scale Data Curation for LLM Training
3) Distributed & Stable LLM Training on Large Clusters
4) Post-training and Evaluation of Pre-trained LLMs
Sessions are scheduled every alternate Wednesday until September 3rd, 2025 (tentatively).
The 1st session focused on hardware and performance where we dive into various communication primitives and determine the gpu topology and in the end, we look into different ways of benchmarking the performance of the cluster using NCCL and MLPerf.
The resource related to the first session can be found here:
https://github.com/ayushbits/llm-deve...
contact [email protected] for any queries
#NPSF #GPU #CDACPune #HPCAI #AI #PARAMSiddhiAI
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
