Hi, I am working on producing a new code review with this new model. Unfortunately, every model is different, even "minor" releases, and I will need some additional time to achieve it. On the upside, the model seems very interesting and could produce more interesting output. Patch review in particular is looking more accurate.
Problems: - The resource use is a lot higher (= each agent will use more context). As a result, I cannot be as aggressive with concurrency, so less tok/s, so review takes significantly longer. - The model will use more expensive tools: it double checks more, verifies its assertions with dedicated tests, and so on. Sometimes it runs Tomcat tests. This puts more pressure on the CPU, which was not happening before. It turns out the new power draw is now too much for my UPS, so a new one is on the way ;) As a result, I expect to have a new core review done by the end of this month, instead of it taking a few hours like before. We'll see if I can still beat the Glasswing stuff ... Rémy --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
