From 565a2f320a2d226c0de921d84a4b6000eb5dfaa1 Mon Sep 17 00:00:00 2001
From: Michael Gschwind <mikekg@meta.com>
Date: Sat, 6 Apr 2024 22:36:33 -0700
Subject: [PATCH] fix minor issues, reduce runtime for macos12 runners

---
 .github/workflows/compile-macos12.yml | 61 +++++++++++++++++++++++++++
 .github/workflows/compile.yml         | 24 +++++------
 .github/workflows/et.yml              | 13 +++---
 model_et.py                           |  2 +-
 4 files changed, 81 insertions(+), 19 deletions(-)
 create mode 100644 .github/workflows/compile-macos12.yml

diff --git a/.github/workflows/compile-macos12.yml b/.github/workflows/compile-macos12.yml
new file mode 100644
index 000000000..d23a48056
--- /dev/null
+++ b/.github/workflows/compile-macos12.yml
@@ -0,0 +1,61 @@
+name: Compile main
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  run-tinystories:
+    strategy:
+      matrix:
+        runner: [macos-12]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v2
+      - name: Setup Python
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.11
+      - name: Print machine info
+        run: |
+          uname -a
+          if [ $(uname -s) == Darwin ]; then
+            sysctl machdep.cpu.brand_string
+            sysctl machdep.cpu.core_count
+          fi
+      - name: Install requirements
+        run: |
+          pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu
+          pip install -r requirements.txt
+      - name: Download checkpoints
+        run: |
+          mkdir -p checkpoints/stories15M
+          pushd checkpoints/stories15M
+          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories15M.pt
+          wget https://github.com/karpathy/llama2.c/raw/master/tokenizer.model
+          popd
+      - name: Run inference
+        run: |          
+          export MODEL_PATH=checkpoints/stories15M/stories15M.pt
+          export MODEL_NAME=stories15M
+          export MODEL_OUT=/tmp
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
+          cat ./output_eager
+          python generate.py --compile --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
+          cat ./output_compiled
+          python export.py --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
+          cat ./output_aoti
+          echo "tests complete"
+          echo "******************************************"
+          # echo "********* EAGER vs TORCH.COMPILE *********"
+          # echo "******************************************"
+          # diff output_eager output_compiled
+          # echo "******************************************"
+          # echo "********* EAGER vs AOT INDUCTOR  *********"
+          # echo "******************************************"
+          # diff output_eager output_aoti
diff --git a/.github/workflows/compile.yml b/.github/workflows/compile.yml
index 88ae1949b..db77c001b 100644
--- a/.github/workflows/compile.yml
+++ b/.github/workflows/compile.yml
@@ -11,7 +11,7 @@ jobs:
   run-tinystories:
     strategy:
       matrix:
-        runner: [ubuntu-latest, macos-12, macos-14]
+        runner: [ubuntu-latest, macos-14]
     runs-on: ${{matrix.runner}}
     steps:
       - name: Checkout repo
@@ -42,13 +42,13 @@ jobs:
         run: |          
           export MODEL_PATH=checkpoints/stories15M/stories15M.pt
           export MODEL_NAME=stories15M
-          export MODEL_DIR=/tmp
+          export MODEL_OUT=/tmp
           python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_eager
           cat ./output_eager
           python generate.py --compile --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
           cat ./output_compiled
-          python export.py --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so  > ./output_aoti
+          python export.py --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
           cat ./output_aoti
 
           echo "******************************************"
@@ -58,8 +58,8 @@ jobs:
           cat ./output_eager
           python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
           cat ./output_compiled
-          python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so  > ./output_aoti
+          python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
           cat ./output_aoti
 
           echo "******************************************"
@@ -69,8 +69,8 @@ jobs:
           cat ./output_eager
           python generate.py --compile --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
           cat ./output_compiled
-          python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so  > ./output_aoti
+          python export.py --quant '{"embedding" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
           cat ./output_aoti
 
           echo "******************************************"
@@ -80,8 +80,8 @@ jobs:
           cat ./output_eager
           python generate.py --compile --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
           cat ./output_compiled
-          python export.py --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so  > ./output_aoti
+          python export.py --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 0}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
           cat ./output_aoti
 
           echo "******************************************"
@@ -91,8 +91,8 @@ jobs:
           cat ./output_eager
           python generate.py --compile --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --temperature 0 > ./output_compiled
           cat ./output_compiled
-          python export.py --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so  > ./output_aoti
+          python export.py --quant '{"linear:int8" : {"bitwidth": 8, "group_size": 8}}' --checkpoint-path ${MODEL_PATH} --output-dso-path ${MODEL_OUT}/${MODEL_NAME}.so
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --dso-path ${MODEL_OUT}/${MODEL_NAME}.so  > ./output_aoti
           cat ./output_aoti
 
           echo "tests complete"
diff --git a/.github/workflows/et.yml b/.github/workflows/et.yml
index 378c6544e..f93cc26d2 100644
--- a/.github/workflows/et.yml
+++ b/.github/workflows/et.yml
@@ -64,12 +64,13 @@ jobs:
         run: |
           export MODEL_PATH=${PWD}/checkpoints/stories15M/stories15M.pt
           export MODEL_NAME=stories15M
+	  export MODEL_OUT=/tmp
+	  
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 > ${MODEL_OUT}/output_eager
+          cat ${MODEL_OUT}/output_eager
 
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 > ${PWD}/output_eager
-          cat ${PWD}/output_eager
-
-          python export.py --checkpoint-path ${MODEL_PATH} --output-pte-path ${PWD}/${MODEL_NAME}.pte
-          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --pte-path ${PWD}/${MODEL_NAME}.pte  > ${PWD}/output_et
-          cat ${PWD}/output_et
+          python export.py --checkpoint-path ${MODEL_PATH} --output-pte-path ${MODEL_OUT}/${MODEL_NAME}.pte
+          python generate.py --checkpoint-path ${MODEL_PATH} --temperature 0 --pte-path ${MODEL_OUT}/${MODEL_NAME}.pte  > ${MODEL_OUT}/output_et
+          cat ${MODEL_OUT}/output_et
 
           echo "Tests complete."
diff --git a/model_et.py b/model_et.py
index a81fe5fcd..6dac4b3da 100644
--- a/model_et.py
+++ b/model_et.py
@@ -11,7 +11,7 @@ def __init__(self, config, path) -> None:
         self.config = config
         self.model_ = exec_lib._load_for_executorch(str(path))
 
-    def forward(self, idxx, input_pos):
+    def forward(self, idx, input_pos):
         # model_.forward expects inputs to be wrapped in a tuple
         forward_inputs = (idx.to(torch.long), input_pos.to(torch.long))
         logits = self.model_.forward(forward_inputs)