Add doc for script

pytorch · Apr 17, 2024 · 114c3d5 · 114c3d5
1 parent 7f88c2d
commit 114c3d5
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 240 deletions.
diff --git a/.ci/scripts/validate.sh b/.ci/scripts/validate.sh
@@ -150,12 +150,16 @@ function generate_aoti_model_output() {
         python -W ignore generate.py --dtype ${DTYPE} --checkpoint-path "$CHECKPOINT_PATH" --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so --device "$TARGET_DEVICE" > "$MODEL_DIR/output_aoti" || exit 1
         cat "$MODEL_DIR/output_aoti"
 
-        # echo "******************************************"
-        # echo "******** INT4 group-wise quantized *******"
-        # echo "******************************************"
-        # python -W ignore export.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path "$CHECKPOINT_PATH" --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so --device "$TARGET_DEVICE" || exit 1
-        # python -W ignore generate.py --dtype ${DTYPE} --checkpoint-path "$CHECKPOINT_PATH" --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so --device "$TARGET_DEVICE" > "$MODEL_DIR/output_aoti" || exit 1
-        # cat "$MODEL_DIR/output_aoti"
+        echo "******************************************"
+        echo "******** INT4 group-wise quantized *******"
+        echo "******************************************"
+        if [ $(uname -s) == "Linux" ]; then
+            echo "Skipping INT4 groupwise quantization because AOTI fails"
+	    else
+            python -W ignore export.py --dtype ${DTYPE} --quant '{"linear:int4" : {"groupsize": 32}}' --checkpoint-path "$CHECKPOINT_PATH" --output-dso-path ${MODEL_DIR}/${MODEL_NAME}.so --device "$TARGET_DEVICE" || exit 1
+            python -W ignore generate.py --dtype ${DTYPE} --checkpoint-path "$CHECKPOINT_PATH" --temperature 0 --dso-path ${MODEL_DIR}/${MODEL_NAME}.so --device "$TARGET_DEVICE" > "$MODEL_DIR/output_aoti" || exit 1
+            cat "$MODEL_DIR/output_aoti"
+        fi
     done
 }
 

diff --git a/.github/workflows/compile-dtype.yml b/.github/workflows/compile-dtype.yml
diff --git a/.github/workflows/compile_t4-dtype.yml b/.github/workflows/compile_t4-dtype.yml
diff --git a/scripts/workflow.sh b/scripts/workflow.sh
@@ -5,6 +5,14 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+################################################################################
+# Usage:
+#   bash script.sh [cpu|cuda] [model_repo] [optional_command]
+# Arguments:
+#   cpu|cuda: Specify the device to run validation on (cpu or cuda).
+#   model_repo: Model repository name to validate (e.g., tinyllamas/stories15M).
+#   optional_command: (optional) Specify additional command "compile", "aoti" or "executorch" to run the selected validation.
+################################################################################
 
 set -eu
 
@@ -75,7 +83,7 @@ MODEL_REPOS=(
     "mistralai/Mistral-7B-v0.1"
     "mistralai/Mistral-7B-Instruct-v0.1"
     "mistralai/Mistral-7B-Instruct-v0.2"
-    # "openlm-research/open_llama_7b"
+    "openlm-research/open_llama_7b"
     "codellama/CodeLlama-7b-Python-hf"
     "codellama/CodeLlama-34b-Python-hf"
     # "meta-llama/Llama-2-7b-chat-hf"