Hipify blog post

--------- Co-authored-by: Suyash Tandon <[email protected]> Co-authored-by: Gina Sitaraman <[email protected]> Co-authored-by: Sriranjani Sitaraman <[email protected]> Co-authored-by: Maria Ruiz Varela <[email protected]> Co-authored-by: Suyash Tandon <[email protected]> Co-authored-by: bobrobey <[email protected]> Co-authored-by: Sam Wu <[email protected]>
ROCm · Apr 26, 2024 · e29e1ac · e29e1ac
1 parent 1ae2a97
commit e29e1ac
Show file tree

Hide file tree

Showing 10 changed files with 832 additions and 24 deletions.
diff --git a/.markdownlint-cli2.yaml b/.markdownlint-cli2.yaml
diff --git a/.markdownlint.yaml b/.markdownlint.yaml
@@ -0,0 +1,18 @@
+default: true
+MD010:
+  code_blocks: false    
+MD013: false
+MD026:
+  punctuation: '.,;:'
+MD029:
+  style: ordered
+MD033: false
+MD034: false
+MD041: false
+MD046: false # Allow indented code blocks (which are output by nbconvert)
+MD051: false
+ignores:
+  - CHANGELOG.md
+  - docs/CHANGELOG.md
+  - "{,docs/}{RELEASE,release}.md"
+  - tools/autotag/templates/**/*.md
diff --git a/blogs/artificial-intelligence/megatron-deepspeed-pretrain/README.md b/blogs/artificial-intelligence/megatron-deepspeed-pretrain/README.md
@@ -92,16 +92,18 @@ you'll use for pre-training. The cell blocks with `%%sh` represent Linux command
 
 ```sh
 %%sh
-#Install DeepSpeed
+python -m pip install --upgrade pip
 
-cd /home/aac
+#Install DeepSpeed
+home_dir=$PWD
+cd $home_dir
 git clone --recursive https://github.com/microsoft/DeepSpeed.git
 cd DeepSpeed
 pip install .[dev,1bit,autotuning]
 
 # Clone Megatron-DeepSpeed
 
-cd /home/aac
+cd $home_dir
 git clone https://github.com/microsoft/Megatron-DeepSpeed.git
 cd Megatron-DeepSpeed
 pip3 install pybind11 nltk transformers
@@ -120,15 +122,15 @@ bash download_vocab.sh
 
 # Pre-process data for oscar dataset
 
-export BASE_SRC_PATH=/home/aac/Megatron-DeepSpeed
+export BASE_SRC_PATH=$home_dir/Megatron-DeepSpeed
 export BASE_DATA_PATH=${BASE_SRC_PATH}/dataset
 python3 ${BASE_SRC_PATH}/tools/preprocess_data.py --input ${BASE_DATA_PATH}/oscar-1GB.jsonl --output-prefix ${BASE_DATA_PATH}/my-gpt2 --vocab-file ${BASE_DATA_PATH}/gpt2-vocab.json --dataset-impl mmap --tokenizer-type GPT2BPETokenizer --merge-file ${BASE_DATA_PATH}/gpt2-merges.txt --append-eod --workers 8
 
 # Install FlashAttention (optional). FlashAttention delivers a rapid and memory-efficient
 # solution for attention mechanisms. If you don't want to use FlashAttention, remove
 # the '--use-flash-attn' flag in the script.
 
-cd /home/aac
+cd $home_dir
 git clone --recursive https://github.com/ROCmSoftwarePlatform/flash-attention.git
 cd flash-attention
 py_version=$(python -V | grep -oP '(?<=[.])\w+(?=[.])')