5

Source document

[2025-04-06 12:35] ajaxdavis: https://thomasalwyndavis--example-axolotl-inference-web.modal.run/?input=what%20is%20resumed%20cli [2025-04-06 12:40] ajaxdavis: https://v0-mood-based-webpage-wbdhsn.vercel.app/ [2025-04-06 12:41] foxhop.: Sup [2025-04-06 12:43] ajaxdavis: (files: image.png) [2025-04-06 12:44] optionalsecurity: (files: image0.png) [2025-04-06 12:44] optionalsecurity: I didn't ask it about a resume though [2025-04-06 12:49] ajaxdavis: using free vercel v0 hosting, just retry [2025-04-06 12:51] ajaxdavis: https://thomasalwyndavis--example-axolotl-inference-web.modal.run/?input=what%20is%20resumed%20cli [2025-04-06 12:51] ajaxdavis: this link should work immediately though [2025-04-06 12:51] ajaxdavis: and just change your question in the url [2025-04-06 12:51] ajaxdavis: but i will probz turn off before bed [2025-04-06 12:51] ajaxdavis: because its running 2 a100's [2025-04-06 13:01] ajaxdavis: `modal-http: Webhook failed: workspace billing cycle spend limit reached` [2025-04-06 13:02] ajaxdavis: https://tenor.com/view/porky-pig-looney-tunes-thats-all-folks-bye-gif-5803466 [2025-04-06 13:03] optionalsecurity: (files: Screenshot_20250406-075039.png) [2025-04-06 13:03] ajaxdavis: aha that's to be expected [2025-04-06 13:04] ajaxdavis: it by nature should only reply to questions about `resumed-cli` because it's an SPM [2025-04-06 13:04] ajaxdavis: hence why that works (files: image.png) [2025-04-06 13:04] ajaxdavis: not bad results for a rickety first pass, the model is about 1.5 gigs [2025-04-06 14:13] lisamegawatts: https://x.com/lioronai/status/1908505039749947617?s=46&t=OS71yaTltL7EclsAy81ZcQ [2025-04-07 01:54] lisamegawatts: Deep dive on technique https://youtu.be/3p0O28W1ZHg?si=AjeDitM7Sz9dt-pZ [2025-04-07 05:33] jonathan.poczatek: what benchmark are you going to use for evals [2025-04-07 05:33] jonathan.poczatek: there are so many on huggingface, thats the part that trips me up [2025-04-07 05:58] traves_theberge: We might have to make our own for this use case. [2025-04-07 06:03] lisamegawatts: Dud they are aelling a dataset of 66 lines for $65 [2025-04-07 06:03] lisamegawatts: I can make datasets [2025-04-07 06:03] lisamegawatts: If they are actually sellable would be cool [2025-04-07 06:04] lisamegawatts: Depends what you want to optimize your model for [2025-04-07 06:04] traves_theberge: We should do that [2025-04-07 06:04] lisamegawatts: I think vp of ai at facebook just resigned over them finetuning llama 4 on benchmarks [2025-04-07 06:05] traves_theberge: dear god. [2025-04-07 06:05] traves_theberge: they did what [2025-04-07 06:05] traves_theberge: https://tenor.com/view/kto-kounotori-kounotoritoken-lbow-storkholders-gif-25676483 [2025-04-07 06:05] lisamegawatts: Gamed the benchmarks, trained on them sirectly like a bunch of cheaters [2025-04-07 06:06] jonathan.poczatek: Everyone does that though [2025-04-07 06:06] jonathan.poczatek: Thats what they are for lol