Yet Another Bookmarks Service



[https://huggingface.co/docs/optimum/index] - - public:mzimmerm
ai, doc, huggingface, llm, model, optimum, repo, small, transformer - 9 | id:1489894 -

Optimum is an extension of Transformers that provides a set of performance optimization tools to train and run models on targeted hardware with maximum efficiency. It is also the repository of small, mini, tiny models.

[https://www.reddit.com/r/LocalLLaMA/comments/12vxxze/most_cost_effective_gpu_for_local_llms/] - - public:mzimmerm
ai, doc, llm, model, optimize, perform - 6 | id:1489804 -

GGML quantized models. They would let you leverage CPU and system RAM, instead of having to rely on a GPU’s. This could save you a fortune, especially if go for some used AMD Epyc platforms. This could be more viable for the larger models, especially the 30B/65B parameters models which would still press or exceed the VRAM on the P40.

[https://medium.com/@andreasmuelder/large-language-models-for-domain-specific-language-generation-how-to-train-your-dragon-0b5360e8ed76] - - public:mzimmerm
ai, article, code, doc, generate, llm, train - 7 | id:1489780 -

training a model like Llama with 2.7 billion parameters outperformed a larger model like Vicuna with 13 billion parameters. Especially when considering resource consumption, this might be a good alternative to using a 7B Foundation model instead of a full-blown ChatGPT. The best price-to-performance base model for our use case turned out to be Mistral 7b. The model is compact enough to fit into an affordable GPU with 24GB VRAM and outperforms the other models with 7B parameters.

Follow Tags