Skip to content

Blackwell local nonrecord#793

Open
pall23-mech wants to merge 6 commits intoopenai:mainfrom
pall23-mech:blackwell-local-nonrecord
Open

Blackwell local nonrecord#793
pall23-mech wants to merge 6 commits intoopenai:mainfrom
pall23-mech:blackwell-local-nonrecord

Conversation

@pall23-mech
Copy link

Adds a non-record submission folder for a local 8 GB Blackwell-class GPU run using train_merged_gpt_flagged.py.
20000 step run, around 19hours runtime.

unpruned packed model: about ~1.21 BPB
final pruned under-cap model: about ~1.25 BPB

Highlights:

  • local constrained-hardware run
  • original packed artifact was slightly over the exact size cap
  • final pruned/repacked artifact fits under the cap
  • includes README and training script

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant