Update README.md

README.md

@@ -10,9 +10,9 @@ Llama-13B converted from official [Llama-13B](https://github.com/facebookresearc
 
 This is updated from [decapoda-research/llama-13b-hf](https://huggingface.co/decapoda-research/Llama-13b-hf) to include (since the many pull requests are not merged yet in decapoda's repo, so I directly open a new repo here):
 
-(1) The naming changes (LLaMA -> Llama) to best fit for `transformers` naming rule, in both `LlamaForCausalLM` and `LlamaTokenizer`. This works perfectly for `transformers>-4.28.0`.
+(1) The naming changes (LLaMA -> Llama) to best fit for `transformers` naming rule, in both `LlamaForCausalLM` and `LlamaTokenizer`. This works perfectly for `transformers>=4.28.0`.
 
-(2) The model checkpoints are saved in 2 shards (instead of 33 shards in [decapoda-research/Llama-13b-hf](https://huggingface.co/decapoda-research/Llama-13b-hf)). Less shards would accelerate loading speed from disk.
+(2) The model checkpoints are saved in 3 shards (instead of 61 shards in [decapoda-research/Llama-13b-hf](https://huggingface.co/decapoda-research/Llama-13b-hf)). Less shards would accelerate loading speed from disk.
 
 --
 license: other