3 ms·
Yeah, JSON mode in Ollama, which isn’t even the full llama.cpp grammar functionality, performs better than OpenAI for me at this point. I don’t understand how t
by bugglebeetle 3y ago
Yeah, JSON mode in Ollama, which isn’t even the full llama.cpp grammar functionality, performs better than OpenAI for me at this point. I don’t understand how they can be raking in billions of dollars and can’t even get this basic stuff right.
- WhitneyLand 3y agollama.cop Performs better than what? 3.5 Turbo, one of the 4.0 models, API or App?
- bugglebeetle 3y agoJSON mode and function-calling with a JSON schema in the OpenAI API.
- WhitneyLand 3y agoRight, but which model? It makes a huge difference.
- bugglebeetle 3y agoI’ve been using OpenChat 3.5 1210 most recently. Before that, Mistral-OpenOrca. Both return JSON more consistently than gpt-3.5-turbo.
- airstrike 3y agogpt-3.5-turbo is not the benchmark
- bugglebeetle 3y agoI don’t know what point you’re trying to make. They also return JSON more consistently than gpt-4, but I don’t use that because it’s overkill and expensive for my text extraction tasks.
- WhitneyLand 3y agoBecause people have different interests and want to hear your results for different reasons. Some want to consider results relative to cost, and some are interested only in how it compares to SOTA.
- bugglebeetle 3y agoI mean, sure, but the parent should also just explicitly state what it is they were asking or claiming. I’ve answered every question asked. Making vague declarations about something not being “the benchmark,” while not stating what you think “the benchmark” should be, is unhelpful.