The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Does anyone here use the Open API responses API?
| Crawly brunch | 08/16/25 |
| Aromatic stock car | 08/16/25 |
| Crawly brunch | 08/16/25 |
| Crawly brunch | 08/16/25 |
| The Barrister of Babylon | 08/17/25 |
| football | 08/17/25 |
Poast new message in this thread
Date: August 16th, 2025 1:31 PM Author: Crawly brunch How do you guys get around the fact that large token blobs get rejected before hitting the api? Responses doesn't allow top level attachments and rejects per message attachments. do you just split everything into 6K token chunks and make a call to extract and then feed it into a final call? I never used this for huge payloads until now, and its pissing me off (http://www.autoadmit.com/thread.php?thread_id=5763083&forum_id=2,#49190248) |
Date: August 16th, 2025 1:53 PM Author: Aromatic stock car This is one reason why Gemini will win in the end
It's only the one that you can start out dumping a huge load of shit on, and then immediately ask extremely specific questions
GPT can do it too but is extremely slow and gets slower and more retarded as more info is accumulated in the same chat |
 |
Date: August 16th, 2025 1:55 PM Author: Crawly brunch im doing it using chunked pipline right now where it summarizes segments at a time and then dumps them into a final extraction adn its still taking forever, its been plodding through this for like 10 min now:
tokens≈21975 (inline_limit 120000) [route] attempting file-attachment path [route] attachments unsupported here; using chunked pipeline [map] summarizing segment 1 [map] summarizing segment 2
slow as fuck its on segment 4 now 5 min later
|
|
Date: August 16th, 2025 1:59 PM Author: Crawly brunch the thing that pisses me off is WHAT THE FUCK IS THE POINT of taking 120K tokens when the nigga endpoint server rjects it before it even hits the NIGGA MODEL?? (http://www.autoadmit.com/thread.php?thread_id=5763083&forum_id=2,#49190293) |
|
Date: August 17th, 2025 10:45 AM Author: The Barrister of Babylon (from the Euphrates to the Forum) I just figured out a way to do it without slow chunking actually using something called ACK mode, where you can send a blob using begin and end syntax and the model acknowledges it got it and then send another doing the same and it synthesizes the two as long as its within the context window which is huge for the new models (http://www.autoadmit.com/thread.php?thread_id=5763083&forum_id=2,#49191759) |
|
Date: August 17th, 2025 11:16 AM Author: football (http://www.autoadmit.com/thread.php?thread_id=5763083&forum_id=2,#49191790) |