19 Dec 2023 |
dashlambda | Now drafting the next mechanism~ | 00:14:56 |
Jern#4673 | Nice, looks great Roger! | 02:33:39 |
| jmansion joined the room. | 12:44:36 |
| guille77 changed their profile picture. | 14:09:16 |
| jarn_62438 joined the room. | 18:44:31 |
20 Dec 2023 |
| clairbee_75188 joined the room. | 13:33:48 |
| clairbee joined the room. | 13:38:53 |
| Apostolos#6329 removed their profile picture. | 15:38:51 |
21 Dec 2023 |
| arv.ai joined the room. | 04:59:09 |
| vtiger_69143 joined the room. | 14:15:06 |
Vortus | Download 411727887_7761146567263004_6255658238765827262_n.mp4 | 14:16:57 |
Vortus |  Download image.png | 14:16:57 |
Vortus | Download birdhouse.py | 14:16:58 |
Vortus | Here's a de-mountable birdhouse made in build123d. Code is a bit hacky to make it in time for the holidays, but some highlights are the auto_finger_joint method at the top of the file, and the dxf exporter at the bottom. Finding a more elegant approach to the spring pull will be a top priority next time I use a compliant mechanism. | 14:16:58 |
Vortus | Download birdhouse.py | 14:42:12 |
aflofo | this was a 10 min side thing but pretty much using LLM's to interface with different CQ libraries like CQ gears to quickly generate stuff with a text interface at a high level prompt. Obviously not for super complex/custom stuff. Thought I'd drop | 19:29:23 |
aflofo | https://github.com/OpenOrion/cqAsk | 19:29:23 |
aflofo |  Download GB4IbLMbQAAioSr.png | 19:29:43 |
Draek#9615 | I've seen this just recently : https://text-to-cad.zoo.dev/ There is interesting things to do with that idea though its better to provide the code it generates as it has little to no value if one cannot edit what's the model outputs ! | 19:40:22 |
Draek#9615 | But great work | 19:40:24 |
aflofo | yeah it allows you to export PY just added that in the export options | 19:40:50 |
Draek#9615 | Is it just bare gpt4 or you fined tune a model to work with CQ? I was just thinking about how we could fine tune a model for generating code cad | 19:41:44 |
aflofo | No just raw GPT, would be intersting to fine tune Llama on a ton of CQ examples but right now I just feed in the API functions for all the parameterizations | 19:42:31 |
aflofo | https://github.com/OpenOrion/CQAsk/blob/main/backend/codex.py#L18 | 19:42:50 |
Draek#9615 | Interesting, if you remove the gear related function can it create one from raw geometry or is it too dumb ? | 19:43:40 |
aflofo | I've done it but not as accurate | 19:44:02 |
aflofo | or consistent | 19:44:05 |
Draek#9615 | Yeah I'd imagine. Fine tuned model could eventually do that I imagine. But this seems to be a very useful end for code cad as it leverage way more easily LLM than trad cad | 19:44:57 |
aflofo | My thinking is LLMs are terrible at ground up CAD but suprising good at putting together high level CAD API parameterization | 19:45:03 |
aflofo | For now | 19:45:08 |