Skip to content

GitLab

Unverified Commit 0c77f878 authored by Shaden Smith's avatar Shaden Smith Committed by GitHub
Browse files

center images (#244)

parent c7d0b0ca
Show whitespace changes
Inline Side-by-side
Showing with 7 additions and 7 deletions
docs/_posts/2020-05-28-fastest-bert-training.md
View file @ 0c77f878
...@@ -57,7 +57,7 @@ practical scenarios range from a few hundred to a few thousand. ...@@ -57,7 +57,7 @@ practical scenarios range from a few hundred to a few thousand.
![Transformer-Kernel-Throughput-128](/assets/images/transformer_kernel_perf_seq128.PNG){: .align-center} ![Transformer-Kernel-Throughput-128](/assets/images/transformer_kernel_perf_seq128.PNG){: .align-center}
![Transformer-Kernel-Throughput-512](/assets/images/transformer_kernel_perf_seq512.PNG) {: .align-center} ![Transformer-Kernel-Throughput-512](/assets/images/transformer_kernel_perf_seq512.PNG){: .align-center}
Figure 1: Performance evaluation of BERT-Large on a single V100 GPU, comparing Figure 1: Performance evaluation of BERT-Large on a single V100 GPU, comparing
DeepSpeed with NVIDIA and HuggingFace versions of BERT in mixed-sequence length DeepSpeed with NVIDIA and HuggingFace versions of BERT in mixed-sequence length
...@@ -102,7 +102,7 @@ approach the GPU peak performance, we employ two lines of optimizations in our ...@@ -102,7 +102,7 @@ approach the GPU peak performance, we employ two lines of optimizations in our
own Transformer kernel implementation: advanced fusion, and invertible own Transformer kernel implementation: advanced fusion, and invertible
operators. operators.
![Transformer-PreLN-Arch](/assets/images/transformer_preln_arch.png) {: .align-center} ![Transformer-PreLN-Arch](/assets/images/transformer_preln_arch.png){: .align-center}
Figure 2: Transformer Layer with Pre-LayerNorm Architecture Figure 2: Transformer Layer with Pre-LayerNorm Architecture
...@@ -133,7 +133,7 @@ shared memory, we reduce the cost of uncoalesced access to main memory to ...@@ -133,7 +133,7 @@ shared memory, we reduce the cost of uncoalesced access to main memory to
better exploit memory bandwidth, resulting in 3% to 5% performance improvement better exploit memory bandwidth, resulting in 3% to 5% performance improvement
in the end-to-end training. in the end-to-end training.
![QKV-Fusion](/assets/images/qkv_fusion.png) {: .align-center} ![QKV-Fusion](/assets/images/qkv_fusion.png){: .align-center}
Figure 3: QKV’s GEMM and transform Kernel-Fusion Figure 3: QKV’s GEMM and transform Kernel-Fusion
...@@ -198,15 +198,15 @@ optimization, we are able to reduce the activation memory of the operator by ...@@ -198,15 +198,15 @@ optimization, we are able to reduce the activation memory of the operator by
half, and the reduced memory allows us to train with larger batch sizes, which half, and the reduced memory allows us to train with larger batch sizes, which
once again improves GPU efficiency. once again improves GPU efficiency.
![Softmax-torch](/assets/images/softmax_pytorch.gif) {: .align-center} ![Softmax-torch](/assets/images/softmax_pytorch.gif){: .align-center}
![Softmax-DS](/assets/images/softmax_deepspeed.gif) {: .align-center} ![Softmax-DS](/assets/images/softmax_deepspeed.gif){: .align-center}
Figure 4: DeepSpeed invertible SoftMax operation versus Default PyTorch SoftMax operation Figure 4: DeepSpeed invertible SoftMax operation versus Default PyTorch SoftMax operation
![LayerNorm-DS](/assets/images/layernorm_pytorch.gif) {: .align-center} ![LayerNorm-DS](/assets/images/layernorm_pytorch.gif){: .align-center}
![LayerNorm-DS](/assets/images/layernorm_deepspeed.gif) {: .align-center} ![LayerNorm-DS](/assets/images/layernorm_deepspeed.gif){: .align-center}
Figure 5: DeepSpeed invertible LayerNorm operation versus Default PyTorch LayerNorm operation Figure 5: DeepSpeed invertible LayerNorm operation versus Default PyTorch LayerNorm operation
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment