this post was submitted on 28 Nov 2023
1 points (100.0% liked)

LocalLLaMA

3 readers
1 users here now

Community to discuss about Llama, the family of large language models created by Meta AI.

founded 1 year ago
MODERATORS
 

I'm looking for what the current best model would be to attempt to do some coding in this realm. realistically, I don't expect a locally run model to be able to help me as much in overall suggestions and a back and forth discussion of what technique to use to solve a particular task. However, I would hope that I can get somehow close, at least in some of this. So the model would need to be fairly well trained on PyTorch and the various machine learning models that are used with that library. Also, some background training on papers that revolve around different machine learning problems and solutions and techniques that are suggested to solve certain things. Which models are currently considered the best for doing any or some of that? The various benchmarks and human eval rubbish just isn't applicable because it seems to be too general. However, I'm open to all suggestions. I understand that 'Phind' is one to look at, but I would love to hear some experience from fellow programmers and system architects on what they found to be useful. My system has an RTX GPU with 24GB and I am in a reluctant way satisfied with one token per second output if it's actually useful output. Thanks in advance for any tips. ( My last post-querying about this seems to have been either blocked or never showed up. Please tell me if I'm doing something wrong.!)

you are viewing a single comment's thread
view the rest of the comments
[–] FullOf_Bad_Ideas@alien.top 1 points 11 months ago

Have you checked deepseek Coder instruct 33b already? I don't know about it's knowledge of pytorch but it's pretty much the best local coding model you can run, so it's your best shot.