A coin flip can make infinite decisions on all possible input taxonomies.
Zero-shot is indeed very cool. And it's super useful/useful for the developer masses that don't know, care, or work pressures don't allow, for proper evaluation and calibration.
But it would be good that we don't over-hype these things.
roughly how much code are you outputting? I'm shocked to hear that API pricing could be lower than subscription. Do you mind sharing how much you're spending monthly on Luna API ?
Are you on the foundation research team for Thomson? (If so, hiya from B!)
Why would you expect Thomson to be particularly good at spam clf? I figured your additional corpus was all news and legal?
I have no connection with Thomson Reuters other than as an end user of a GGUF of the LLM I mentioned. That said, from my personal experience with this specific LLM, it's a decent improvement over a "base" Qwen 3.6 35B A3B Q8, and it does a good job of analyzing and categorizing documents on relatively small resources. It'll run fine in llama-server in pure CPU only on a 64GB RAM system with plenty of room to spare, takes something like 47GB with RAM reserved in llama-server for cache and full context size.
This is beautiful and I'm now itching to make a workbench (finally! that initial "how many sheets and what for what size" problem has been the mental blocker)
Glad ShopSpec helped get you over the hump and appreciate the feedback. I am building a tool that I find useful for myself and previously encountered a similar pattern to what you mention.
Zero-shot is indeed very cool. And it's super useful/useful for the developer masses that don't know, care, or work pressures don't allow, for proper evaluation and calibration.
But it would be good that we don't over-hype these things.