From b3c2d9a8e7ae379a028ca4d53204df32a912e5bc Mon Sep 17 00:00:00 2001 From: Aydyn Tairov Date: Mon, 7 Aug 2023 12:58:40 +0100 Subject: [PATCH] Fix README --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 1d5cd8d5..97fe664b 100644 --- a/README.md +++ b/README.md @@ -5,9 +5,9 @@ I think it's worthwhile to have a fork that implements Python version of `llama2.c` as a kind of simplest "reference implementation" for further educational purpose for wide audience. -Currently `llam2.c` contains 2`.py` files for training models and and 1 `.c` file for inference. There is a gap to have a simple reference implementation of all transformers logic in a simple Python file under 500 lines. While original FB/llama code also implemented in Python, I couldn't say that implementation is simple due to multiple dependencies and optimizations impelemnted. +Currently `llam2.c` contains 2`.py` files for training models and 1 `.c` file for inference. There is a gap to have a simple reference implementation of all transformers logic in a simple Python file under 500 lines. While original FB/llama code also implemented in Python, I couldn't say that implementation is simple due to multiple dependencies and optimizations impelemnted. -PS. Performance is awful at the moment `~1 tok./sec`, there is a huge room for improvements. PRs are welcome 😄 +PS. Performance is awful at the moment `~1 tok./sec`, there is a huge room for improvements ## feel the magic