Uploaded image for project: 'Performance and Scale for AI Platforms'
  1. Performance and Scale for AI Platforms
  2. PSAP-1468

LLM FSDP training

XMLWordPrintable

    • Icon: Epic Epic
    • Resolution: Done
    • Icon: Major Major
    • None
    • None
    • None
    • None
    •  LLM FSDP training
    • RHOAI, Training
    • Not Selected
    • False
    • False
    • None
    • 0% To Do, 0% In Progress, 100% Done

      now that we have multi-node training working, let's use FSDP and see what are the benefits of offloading some work to the CPU. 

      for this training session, I will start with the granite-7b module  which is shipped with RHEL  and in case thing wont work out might switch to Meta-Llama-3-8B

          1.
          		 Get FSDP to work Sub-task Closed Undefined Ashish Kamra

              bbenshab Boaz Ben Shabat
              bbenshab Boaz Ben Shabat
              Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

                Created:
                Updated:
                Resolved: