Tostino

30 ms·

751.
by Tostino 3y ago
Yeah, check out some of the showcase I posted above with some more info: https://news.ycombinator.com/item?id=38482347
752.
by Tostino 3y ago
Absolutely, pick a complicated problem and keep breaking it down with an existing model (whatever sota) until you have a consistent output for each step of your problem. And then stitch all the outputs together into a coherent single respon
753.
by Tostino 3y ago
Lots and lots of manual review of very detailed instructions to "more powerful LLMs" with 2-4 prompts to generate the training data.
754.
by Tostino 3y ago
I'm not competing with OpenAI... I did a whole bunch of work, and released it for anyone who wants to use it. It does what I trained it on well. Use it if you want to, or don't. Either way.
755.
by Tostino 3y ago
Try out my model vs gpt4 for the same tasks (I explicitly trained on) and compare. https://huggingface.co/Tostino/Inkbot-13B-8k-0.2 It's a 13b param model that isn't meant to be general purpose, but is meant
756.
by Tostino 3y ago
Sure thing, I used axolotl, and my training parameters were: sequence_len: 6144 lora_r: 128 lora_alpha: 48 learning_rate: 0.00006 warmup_steps: 600 lr_scheduler: cosine gradient_accumulation_steps: 4 micro_batch_size: 1 num_epochs: 4 optimi
757.
by Tostino 3y ago
It depends on what you're doing... Just for reference, here is a small showcase of the capabilities that I've trained on a 13 billion parameter llama2 fine tune (done with qlora). https://old.reddit.com/r/Loca
758.
by Tostino 3y ago
It's super easy to replace an OpenAI api endpoint with an Azure api endpoint. You totally correct here. I don't see why people are acting like this is a risk at all.
759.
by Tostino 3y ago
There are open ai compatible chat/completion endpoints for local LLMs. You point the url to your self hosted version, and use the API key you started it with...
760.
by Tostino 3y ago
I'm doing that right now. It works, but tethering limits are definitely enforced.
761.
by Tostino 3y ago
Different person, but just about everything. It simply wasn't enjoyable, at all. I'm just talking about the first hobbit, I have somehow managed to purge the memory of seeing the others from my memory.
762.
by Tostino 3y ago
Use actual evidence for your argument. You claimed ignorance, why should I believe anything you say?
763.
by Tostino 3y ago
Again, it's possible because of volume.
764.
by Tostino 3y ago
The rising sea levels aren't about more water being in the environment, it's about more water being in liquid form rather than stored in solid form in glaciers on land. The small amount of water being in a tree trunk vs released b
765.
by Tostino 3y ago
That's...not how this works at all. Is this a joke?
766.
by Tostino 3y ago
I'd agree that what is available publicly isn't anything that hasn't been in wide discussion for an agent framework since maybe ~march/april of this year, and many people had just hacked together their own version with a
767.
by Tostino 3y ago
The fake data that's been caught so far in just about every one of the cases that I've read about has been ridiculously poorly constructed fake data. I think the people that are good at faking data simply don't get caught.
768.
by Tostino 3y ago
So explain yourself in that case, if it should be so obvious, it shouldn't take long.
769.
by Tostino 3y ago
The output is a pattern for a specific input* They need to know the input...
770.
by Tostino 3y ago
Most laptops come with some software bundled that helps configure this. Dell does I know. So did Lenovo.
771.
by Tostino 3y ago
Correct.
772.
by Tostino 3y ago
No, like the other comment said, it's just using the `n` parameter in an OpenAI style API. For example, vLLM and llamacpp have support for it.
773.
by Tostino 3y ago
Right, most inference servers support this already.
774.
by Tostino 3y ago
What a total mess this has been all around.
775.
by Tostino 3y ago
Yup, switching to sequential uuids reduced the wal traffic in my SaaS database significantly when we did it ~6 years ago.
776.
by Tostino 3y ago
Exactly what I was trying to say. Thank you. Still works best if you know what you are doing and can give very detailed instructions, but GPT-4 is vastly more capable than Bing for these tasks.
777.
by Tostino 3y ago
We were talking about code generation though. It's horrible at it.
778.
by Tostino 3y ago
Honestly I've had it guide me to better practices (like described) on side projects than I would have implemented on my own.
779.
by Tostino 3y ago
Bing is far and away worse than gpt4 through ChatGPT or the API, just FYI. Don't even consider it comparable, even if they say it is the same model under the hood. Their "optimizations" have crippled it's capabilities if
780.
by Tostino 3y ago
That was actually part of how I intended my "negotiated" to be taken. Part of it is a monetary negotiation where we invest more in base load and peak load, but some of it needs to be in the ability for the grid to request to shed

More ›