use correct fsdp act ckpt & redo benchmark (#32)

README.md

@@ -12,7 +12,7 @@
 
 | **Example**                                    | **Description**                                                                                   | **Lightning Studio** |
 |------------------------------------------------|---------------------------------------------------------------------------------------------------|----------------------|
-| [**Hugging Face Trainer**](https://github.com/linkedin/Liger-Kernel/tree/main/examples/huggingface)      | Increase 20% throughput and reduce memory usage by 60% with LLaMA3-8B on Alpaca dataset using 4 A100s | TBA                  |
+| [**Hugging Face Trainer**](https://github.com/linkedin/Liger-Kernel/tree/main/examples/huggingface)      | Train llama3 8B ~20% faster with over 40% memory reduction on Alpaca dataset using 4 A100s | TBA                  |
 | [**Lightning Trainer**](https://github.com/linkedin/Liger-Kernel/tree/main/examples/lightning)         | Increase 15% throughput and reduce memory usage by 40% with LLaMA3-8B on MMLU dataset using 8 A100s  | TBA                  |
 
 ### Advanced

examples/huggingface/README.md

@@ -14,5 +14,10 @@ sh run.sh
 
 
 ## Benchmark Result
-![MFU](img/hf_mfu.png)
+
+Benchmark conditions: LLaMA 3-8B, Alpaca Dataset, Max seq len = 512, Data Type = bf16, Optimizer = AdamW, Gradient Checkpointing = True, Distributed Strategy = FSDP1 on 4 A100s.
+
+The throughput increases by approximately 20% with more data, but the GPU memory is reduced by 40%. This means you can train the model on smaller GPUs, with larger batch sizes, or with longer sequence lengths at no additional cost.
+
+![Throughput](img/hf_tps.png)
 ![GPU Memory Allocated](img/hf_mem_alloc.png)