Skip to content

Apply driving dataset limits while loading - #48

Closed
haraschax wants to merge 1 commit into
mainfrom
agent/training-list-limit
Closed

Apply driving dataset limits while loading#48
haraschax wants to merge 1 commit into
mainfrom
agent/training-list-limit

Conversation

@haraschax

@haraschax haraschax commented Jul 30, 2026

Copy link
Copy Markdown

What

  • accept SI-formatted driving dataset limits such as 100k and 1M
  • pass the worldmodel limit to the shared route-list reader before splitting and shuffling
  • keep the path trainer's limit compatible with the same CLI syntax

Why

Applying a limit after loading and shuffling still parses the entire route list and does not select its requested prefix.

Impact

Driving dataset limits now select the prefix of the configured list. Other TorchTitan experiments are unchanged.

Dependency

Paired with commaai/xx#6892, which adds the shared prefix-loading API and pins this commit.

Checks

  • required pre-commit suite on all changed files
  • path and worldmodel CLI parsing with 100k / 1M
  • worldmodel loader check verifying the limit is forwarded while reading

@haraschax
haraschax force-pushed the agent/training-list-limit branch 2 times, most recently from f422f36 to a03363a Compare July 30, 2026 19:18
@haraschax
haraschax force-pushed the agent/training-list-limit branch from a03363a to 7c8a3c9 Compare July 31, 2026 00:02
@haraschax haraschax closed this Aug 21, 2026
@haraschax
haraschax deleted the agent/training-list-limit branch August 21, 2026 22:26
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