Richard Feldman created
This is based on having observed that there is a lot of variation
between runs on `n=1` and `n=3`.
* With `n=8` two runs on the same branch give answers that seem close
enough to be reasonably consistent.
* With higher concurrency, trying to run this many repetitions seems to
lead language servers to time out a lot, causing evals to fail.
Release Notes:
- N/A.github/workflows/eval.yml | 2 +-
crates/eval/src/eval.rs | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
@@ -69,7 +69,7 @@ jobs:
run: cargo build --package=eval
- name: Run eval
- run: cargo run --package=eval -- --repetitions=3 --concurrency=1
+ run: cargo run --package=eval -- --repetitions=8 --concurrency=1
# Even the Linux runner is not stateful, in theory there is no need to do this cleanup.
# But, to avoid potential issues in the future if we choose to use a stateful Linux runner and forget to add code
@@ -52,10 +52,10 @@ struct Args {
#[arg(long, value_delimiter = ',', default_value = "rs,ts")]
languages: Vec<String>,
/// How many times to run each example.
- #[arg(long, default_value = "1")]
+ #[arg(long, default_value = "8")]
repetitions: usize,
/// Maximum number of examples to run concurrently.
- #[arg(long, default_value = "10")]
+ #[arg(long, default_value = "4")]
concurrency: usize,
}