How LLMs Work, Explained Without Math - miguelgrinberg.com

[https://blog.miguelgrinberg.com/post/how-llms-work-explained-without-math] - 2024-05-13 18:48:29 - public:xxx

ai, algorithms, artificial, code, explained, intelligence, learn, llama, llm - 9 | id:1492049 -

(2) Are there any tiny (1-3b) models finetuned for coding available in GGUF format? : LocalLLaMA

[https://www.reddit.com/r/LocalLLaMA/comments/16csdq6/are_there_any_tiny_13b_models_finetuned_for/] - 2024-03-04 10:56:19 - public:mzimmerm

ai, code, generate, llm, model, newspeak, small - 7 | id:1489789 -

bigcode (BigCode)

[https://huggingface.co/bigcode] - 2024-03-04 10:50:02 - public:mzimmerm

ai, code, generate, huggingface, llm, model, newspeak, santacoder, small, starcoder - 10 | id:1489788 -

Research community developing various code models, small and big. Models may not be instruct

WizardLM (WizardLM)

[https://huggingface.co/WizardLM] - 2024-03-04 10:42:44 - public:mzimmerm

ai, code, generate, huggingface, llm, model, newspeak, small, wizardcoder - 9 | id:1489787 -

Another open source small (1B) model.

deepseek-ai (DeepSeek)

[https://huggingface.co/deepseek-ai] - 2024-03-04 10:24:32 - public:mzimmerm

ai, best, code, deepseek, good, huggingface, instruct, llm, model, newspeak, small - 11 | id:1489786 -

They have the 1.3B version!!! This may be the best to start with Newspeak. Should work train even on huggingcface

deepseek-ai/deepseek-coder-6.7b-instruct · Hugging Face

[https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct] - 2024-03-04 10:13:20 - public:mzimmerm

ai, code, generate, good, llm, model, newspeak, opensource - 8 | id:1489783 -

Another possible model. For coding capabilities, Deepseek Coder achieves state-of-the-art performance among open-source code models on multiple programming languages and various benchmarks.

LLaMA 7B GPU Memory Requirement - Transformers - Hugging Face Forums

[https://discuss.huggingface.co/t/llama-7b-gpu-memory-requirement/34323/6] - 2024-03-04 10:10:38 - public:mzimmerm

ai, code, generate, llama, llm, model, newspeak, train - 8 | id:1489782 -

With the optimizers of bitsandbytes (like 8 bit AdamW), you would need 2 bytes per parameter, or 14 GB of GPU memory.

stabilityai/stable-code-3b · Hugging Face

[https://huggingface.co/stabilityai/stable-code-3b] - 2024-03-04 10:05:36 - public:mzimmerm

ai, code, generate, llm, model, newspeak - 6 | id:1489781 -

Another potential model to use for Newspeak, but it is NOT open source. Adventage: 2.5B params, so should be usable in small GPUs

Large Language Models for Domain-Specific Language Generation: How to Train Your Dragon | by Andreas Mülder | Medium

[https://medium.com/@andreasmuelder/large-language-models-for-domain-specific-language-generation-how-to-train-your-dragon-0b5360e8ed76] - 2024-03-04 09:45:59 - 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.