Repository navigation
feat(config): 16k output tokens by default (plan #12); full run 0.539 - #23
Merged
Merged
Conversation
With the act nudge, 16k reached 19/30 on the dev slice against 14/30 at 8k, and cut-off give-ups fell from 8 to 3. Turns are slower (46 vs 26 s), and timeouts rose from 2 to 6.
rzhang569
changed the title
Collaborator
Author
|
This was merged into |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #22 (which is stacked on #21).
LLM_MAX_TOKENS8192 → 16384, in.env.opand as the code default. Update your own.env.op.local, it still says 8192.On the 10-task slice with #22's act nudge, 3 trials:
Full run 2 with both (#22 + this): 0.539, 48/89, up from 0.438 (39/89). Give-ups 25 → 11; timeouts 9 → 20, now the biggest failure. Data in
eval/results/full-run-2.csv, details in docs/experiments.md.