4294967295

[doublemover]

Good work on the columns but the values are still getting capped

[doublemover]

I don't think you got this one right, needs bigger ints. Using 3.3.4

[mike1858]

@doublemover Ah yes, I forgot to release. The fix itself works, but I just never made a release. I'll be creating it now.

[Sewer56]

What the-
What on earth are your stats?

I was really, really generous when I set the limits as I was balancing memory usage; and this one just blows them out of the water.

How in the holy earth is this dude breaking >4 billion tokens a day on the regular; is this a SaaS' collected usage or something?
Best I ever got was around 280 million, by leaving 5 runners on for around 8 hours.

Anyway @mike1858 , you forgot to update the comment, still reads

/// Lightweight stats for TUI display only (24 bytes vs 320 bytes for full Stats).

It's 48 bytes now.

On that note. The cost_cents does not need widening here either. This wastes 8 bytes; once padding is also accounted for.
So this could be 40 bytes, with cost_cents still as u32.
The TuiStats only aggregates for Session/Message/Day

If this person manages to break 43 million dollars of expenditure a day, I'll be damned...
They only got 177 thousand in 3 months; so not anywhere close, so cost_cents should stay a u32.

[mike1858]

Released v3.3.5.

Tip

In your ~/.splitrail.toml (Windows: %USERPROFILE%\.splitrail.toml), you can set number_human to true and number_comma to true. That way, instead of numbers appearing as 616158035, they'll appear as 616.15m and for costs, they'll show up as $71,085.09. Because you have huge numbers, and that could make them a lot easier to read.

[doublemover]

I'm just one dude with one pro acct. 4B/day is easy to clear, with the near-daily limit resets I've been aiming for using the whole week ASAP, which is ~9B with the 2x'd limit (the 12th is a good example of this).

I've been cooking on PairOfCleats, which is a private local-first alternative to Sourcegraph with support for 20+ langs/frameworks, and Slopjective-C 3.0, which is just "Objective-C 3.0" as I've always wanted it. And a couple games. So two to five 5.3-codex and more recently 5.4 high/xhighs, each using 6 sub agents.

My current limit is local compute, the PoC benchmark indexes 50M lines of code across a few hundred repos and I can do that ~twice a day alongside the other projects (these values are skewed by smaller repos/aborted runs). Any more than that and I find the limits of 128GB ram/storage io. I started both projects with the intention of "hey this is way too much work lets see how long it'll take"

Can I borrow like $5,000,000 I wanna try something 💀

Now that I can see the full values, it appears that you are not counting tokens correctly (every other utility also does this, so don't even worry about it). Haven't looked into it at all, unsure if each TUI update is creating duplicate session files or what, the sessionid's should match across them. Sessions that span multiple days are likely also causing some insane values. Seems to also be confusing the fuck out of OAI while they track down their usage drain issues 😭 not to throw shade but I figured they would be doing a little better, given they have unlimited tokens.

Ty for update tho!! I have not been uploading to the leaderboard, some of these sessions are >1 gig at this point lmao

[Sewer56]

Damn. Hella crazy.