From 79028b14b3d4c094f0e865da01e39441e78928d7 Mon Sep 17 00:00:00 2001 From: Hamid Shojanazeri <hamid.nazeri2010@gmail.com> Date: Tue, 18 Jul 2023 23:48:20 +0000 Subject: [PATCH] update notes for int8 lack of support in FSDP --- README.md | 2 +- docs/mutli_gpu.md | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index ff6adee3..90800bcd 100644 --- a/README.md +++ b/README.md @@ -81,7 +81,7 @@ Here we make use of Parameter Efficient Methods (PEFT) as described in the next ### Multiple GPUs One Node: -**NOTE** please make sure to use PyTorch Nightlies for using PEFT+FSDP . +**NOTE** please make sure to use PyTorch Nightlies for using PEFT+FSDP. Also, note that int8 quantization from bit&bytes currently is not supported in FSDP. ```bash diff --git a/docs/mutli_gpu.md b/docs/mutli_gpu.md index b0ca2e9f..5695ccf5 100644 --- a/docs/mutli_gpu.md +++ b/docs/mutli_gpu.md @@ -26,6 +26,8 @@ This runs with the `samsum_dataset` for summarization application by default. **Multiple GPUs one node**: +**NOTE** please make sure to use PyTorch Nightlies for using PEFT+FSDP. Also, note that int8 quantization from bit&bytes currently is not supported in FSDP. + ```bash torchrun --nnodes 1 --nproc_per_node 4 ../llama_finetuning.py --enable_fsdp --model_name /patht_of_model_folder/7B --use_peft --peft_method lora --output_dir Path/to/save/PEFT/model -- GitLab