[Doc][Example] Fine-tune vicuna-13b-v1.3 with LightningTrainer + De…

…epSpeed (ray-project#37016) Signed-off-by: woshiyyya <xiaoyunxuan1998@gmail.com> Signed-off-by: Yunxuan Xiao <xiaoyunxuan1998@gmail.com> Signed-off-by: matthewdeng <matthew.j.deng@gmail.com> Co-authored-by: matthewdeng <matthew.j.deng@gmail.com> Signed-off-by: e428265 <arvind.chandramouli@lmco.com>
lmco · Aug 31, 2023 · 7a9f3e0 · 7a9f3e0
1 parent b20e982
commit 7a9f3e0
Show file tree

Hide file tree

Showing 10 changed files with 1,515 additions and 1 deletion.
diff --git a/doc/source/_toc.yml b/doc/source/_toc.yml
@@ -139,6 +139,8 @@ parts:
                 title: "Torch Data Prefetching Benchmark"
               - file: train/examples/pytorch/pytorch_resnet_finetune
                 title: "PyTorch Finetuning ResNet Example"
+              - file: train/examples/lightning/vicuna_13b_lightning_deepspeed_finetune
+                title: "Fine-tune Vicuna-13B with DeepSpeed and PyTorch Lightning"
           - file: train/faq
           - file: train/api/api
 

diff --git a/doc/source/ray-overview/examples.rst b/doc/source/ray-overview/examples.rst
@@ -1618,3 +1618,10 @@ Ray Examples
         .. button-ref:: /serve/tutorials/streaming
 
             Using Ray Serve to deploy a chatbot
+
+    .. grid-item-card:: :bdg-secondary:`Code example`
+        :class-item: gallery-item training llm
+
+        .. button-ref:: /train/examples/lightning/vicuna_13b_lightning_deepspeed_finetune
+
+            Fine-tune vicuna-13b-v1.3 with DeepSpeed and LightningTrainer
diff --git a/doc/source/train/examples.rst b/doc/source/train/examples.rst
@@ -72,6 +72,14 @@ Distributed Training Examples using Ray Train
         .. button-ref:: dolly_lightning_fsdp_finetuning
 
             Fine-tune LLM with AIR LightningTrainer and FSDP
+
+    .. grid-item-card::
+        :img-top: /images/pytorch_lightning_small.png
+        :class-img-top: pt-2 w-75 d-block mx-auto fixed-height-img
+
+        .. button-ref:: vicuna_lightning_deepspeed_finetuning
+
+            Fine-tune vicuna-13b-v1.3 with Deepspeed and LightningTrainer
 
 
 Ray Train Examples Using Loggers & Callbacks

diff --git a/doc/source/train/examples/lightning/BUILD b/doc/source/train/examples/lightning/BUILD
@@ -10,7 +10,10 @@ filegroup(
 py_test_run_all_notebooks(
     size="large",
     include=["*.ipynb"],
-    exclude=["lightning_exp_tracking.ipynb"],
+    exclude=[
+        "lightning_exp_tracking.ipynb", # CPU test
+        "vicuna_13b_lightning_deepspeed_finetune.ipynb", # Release Test
+    ],
     data=["//doc/source/train/examples/lightning:lightning_examples"],
     tags=["exclusive", "team:ml", "gpu", "ray_air"],
 )

diff --git a/doc/source/train/examples/lightning/vicuna_13b_lightning_deepspeed_finetune.ipynb b/doc/source/train/examples/lightning/vicuna_13b_lightning_deepspeed_finetune.ipynb
diff --git a/release/air_examples/vicuna_13b_lightning_deepspeed_finetuning/test_myst_doc.py b/release/air_examples/vicuna_13b_lightning_deepspeed_finetuning/test_myst_doc.py
@@ -0,0 +1 @@
+../../../doc/test_myst_doc.py
diff --git a/..._examples/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_deepspeed_compute_aws.yaml b/..._examples/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_deepspeed_compute_aws.yaml
@@ -0,0 +1,20 @@
+cloud_id: {{env["ANYSCALE_CLOUD_ID"]}}
+region: us-west-2
+
+head_node_type:
+    name: head_node
+    instance_type: g5.16xlarge
+
+worker_node_types:
+    - name: worker_node
+      instance_type: g5.4xlarge
+      min_workers: 15
+      max_workers: 15
+      use_spot: false
+
+aws:
+  TagSpecifications:
+    - ResourceType: "instance"
+      Tags:
+        - Key: ttl-hours
+          Value: '24'
diff --git a/release/air_examples/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_deepspeed_env.yaml b/release/air_examples/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_deepspeed_env.yaml
@@ -0,0 +1,27 @@
+base_image: {{ env["RAY_IMAGE_ML_NIGHTLY_GPU"] | default("anyscale/ray:nightly-py38-cu118") }}
+env_vars: {}
+debian_packages:
+  - curl
+
+python:
+  pip_packages:
+    - datasets==2.13.1
+    - evaluate==0.4.0
+    - scikit-learn==1.3.0
+    - boto3==1.28.5
+    - myst-parser==0.15.2
+    - myst-nb==0.13.1
+    - jupytext==1.13.6
+    - typing-extensions<4.6.0
+  conda_packages: []
+
+post_build_cmds:
+  - pip uninstall -y ray || true && pip3 install -U {{ env["RAY_WHEELS"] | default("ray") }}
+  - {{ env["RAY_WHEELS_SANITY_CHECK"] | default("echo No Ray wheels sanity check") }}
+  - echo "sudo lsblk -f" >> ~/.bashrc
+  - echo "yes N | sudo mkfs -t ext4 /dev/nvme1n1 || true" >> ~/.bashrc
+  - echo "mkdir -p /mnt/local_storage" >> ~/.bashrc
+  - echo "sudo chmod 0777 /mnt/local_storage" >> ~/.bashrc
+  - echo "sudo mount /dev/nvme1n1 /mnt/local_storage || true" >> ~/.bashrc
+  - pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
+  - pip3 install "pytorch_lightning==2.0.3" "transformers==4.30.2" "accelerate==0.20.3" "deepspeed==0.9.4"
diff --git a/...s/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_lightning_deepspeed_finetune.ipynb b/...s/vicuna_13b_lightning_deepspeed_finetuning/vicuna_13b_lightning_deepspeed_finetune.ipynb
@@ -0,0 +1 @@
+../../../doc/source/train/examples/lightning/vicuna_13b_lightning_deepspeed_finetune.ipynb
diff --git a/release/release_tests.yaml b/release/release_tests.yaml
@@ -967,6 +967,27 @@
 
   # variations: TODO(jungong): add GCP variation.
 
+- name: air_example_vicuna_13b_lightning_deepspeed_finetuning
+  group: AIR examples
+  working_dir: air_examples/vicuna_13b_lightning_deepspeed_finetuning
+
+  python: "3.8"
+
+  frequency: weekly
+  team: ml
+  cluster:
+    byod:
+      type: cu118
+      pip:
+        - myst-parser==0.15.2
+        - myst-nb==0.13.1
+        - jupytext==1.13.6
+    cluster_env: vicuna_13b_deepspeed_env.yaml
+    cluster_compute: vicuna_13b_deepspeed_compute_aws.yaml
+
+  run:
+    timeout: 4700
+    script: python test_myst_doc.py --path vicuna_13b_lightning_deepspeed_finetune.ipynb
 
 #####################################
 # Workspace templates release tests #