The optimizations bring cumulative memory savings to about 25%.
Boaz Barak#140
Since my last post, I reduced ty’s retained memory by another 15% with Codex. We're now at a ~25% memory reduction overall via /goal, largely in the background.
I love working with the GPT models so much.
Positive users highlight the impressive 25% memory reduction in Ty using AI tools as a valuable background improvement, while negative users complain about usage limits and billing issues with Codex.
Playing codex like an orchestra. One /goal at a time.
Since my last post, I reduced ty’s retained memory by another 15% with Codex. We're now at a ~25% memory reduction overall via /goal, largely in the background.
I love working with the GPT models so much.
Bang
https://github.com/astral-sh/ruff/pull/25501 https://github.com/astral-sh/ruff/pull/25593 https://github.com/astral-sh/ruff/pull/25606 https://github.com/astral-sh/ruff/pull/25609 https://github.com/astral-sh/ruff/pull/25610 https://github.com/astral-sh/ruff/pull/25649 https://github.com/astral-sh/ruff/pull/25656 https://github.com/astral-sh/ruff/pull/25660 https://github.com/astral-sh/ruff/pull/25661 https://github.com/astral-sh/ruff/pull/25663
Bang
https://github.com/astral-sh/ruff/pull/25669 https://github.com/astral-sh/ruff/pull/25675 https://github.com/astral-sh/ruff/pull/25682 https://github.com/astral-sh/ruff/pull/25689 https://github.com/astral-sh/ruff/pull/25692 https://github.com/astral-sh/ruff/pull/25694 https://github.com/astral-sh/ruff/pull/25701 https://github.com/astral-sh/ruff/pull/25710 https://github.com/astral-sh/ruff/pull/25723 https://github.com/astral-sh/ruff/pull/25735
@charliermarsh Have you tried "these are the fixes we've done before, are there any other instances of this across the codebase"? Also at this point a skill might be helpful to document knowledge from past PRs
@charliermarsh @jxnlco Soon ty won’t even use the cpu because too optimized 😂
@charliermarsh wow, he (& codex) just keeps shipping!!
@charliermarsh this is the real AI coding story nobody tweets about. not vibe-coded demos, love to see it🫡
@charliermarsh Do you attribute these improvements to Codex itself, the ability to iterate faster, or something else?
@charliermarsh please, implement something in UV ecosystem to distribute python apps like pyapp or something which can replace pyinstaller. It is a nightmare :(
@thsottiaux Can the 5.5 pro model be used from codex? I'm making a simulator and wish the deep research stuff and codex were better connected? I'm also new to codex so might be missing something.
@charliermarsh So it is expected that /goal takes a while to run
@charliermarsh impressive!
@charliermarsh Which model ? 5.5 ?
@thsottiaux
@thsottiaux "/goal" in codex is basically the pinnacle of ai. properly harnessed, it can solve literally any problem and it will keep going until it is done.
@thsottiaux
@thsottiaux If Codex is the orchestra, that 25% memory reduction is the standing ovation. Honestly impressive how much of this happened in the background. What's the next movement?
We were spoiled by the 20x limits and then suddenly they disappeared 😭Now I burn through my weekly quota in 2 days. When doctors taper people off medication, they don't go from full dose to nothing overnight. Atleast wean off responsibly like full dose, half dose, 1/4 etc. Withdrawal symptoms from losing codex are real 😅💊
CC: @sama 🫡
@thsottiaux The biggest mistake with Codex is treating it like a chatbot.
The magic happens when you give it one clear goal at a time and let it execute.
Conduct, don’t micromanage.
@charliermarsh the modern day saint
The optimizations bring cumulative memory savings to about 25%.
Boaz Barak#140Since my last post, I reduced ty’s retained memory by another 15% with Codex. We're now at a ~25% memory reduction overall via /goal, largely in the background.
I love working with the GPT models so much.