From 66c2e45634de96959b363b3464c4b908febdfd68 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cheen=20Hau=2C=20=E4=BF=8A=E8=B1=AA?=
 <33478814+chtanch@users.noreply.github.com>
Date: Tue, 17 Oct 2023 14:46:41 +0800
Subject: [PATCH] Add unit tests for optimized model correctness (#9151)

* Add test to check correctness of optimized model

* Refactor optimized model test

* Use models in llm-unit-test

* Use AutoTokenizer for bloom

* Print out each passed test

* Remove unused tokenizer from import
---
 .../llm/test/inference/test_optimize_model.py | 58 +++++++++++++++++++
 python/llm/test/run-llm-inference-tests.sh    |  4 +-
 2 files changed, 60 insertions(+), 2 deletions(-)
 create mode 100644 python/llm/test/inference/test_optimize_model.py

diff --git a/python/llm/test/inference/test_optimize_model.py b/python/llm/test/inference/test_optimize_model.py
new file mode 100644
index 00000000..6d593e72
--- /dev/null
+++ b/python/llm/test/inference/test_optimize_model.py
@@ -0,0 +1,58 @@
+#
+# Copyright 2016 The BigDL Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import pytest
+import os
+
+from bigdl.llm.transformers import AutoModelForCausalLM, AutoModel
+from transformers import LlamaTokenizer, AutoTokenizer
+
+
+llama_model_path = os.environ.get('LLAMA_ORIGIN_PATH')
+bloom_model_path = os.environ.get('BLOOM_ORIGIN_PATH')
+chatglm2_6b_model_path = os.environ.get('ORIGINAL_CHATGLM2_6B_PATH')
+replit_code_model_path = os.environ.get('ORIGINAL_REPLIT_CODE_PATH')
+
+prompt = "Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun"
+
+@pytest.mark.parametrize("Model, Tokenizer, model_path, prompt", [
+    (AutoModelForCausalLM, LlamaTokenizer, llama_model_path, prompt),
+    (AutoModelForCausalLM, AutoTokenizer, bloom_model_path, prompt),
+    (AutoModel, AutoTokenizer, chatglm2_6b_model_path, prompt),
+    (AutoModelForCausalLM, AutoTokenizer, replit_code_model_path, prompt),
+])
+def test_optimize_model(Model, Tokenizer, model_path, prompt):
+    tokenizer = Tokenizer.from_pretrained(model_path, trust_remote_code=True)
+    input_ids = tokenizer.encode(prompt, return_tensors="pt")
+
+    model = Model.from_pretrained(model_path,
+                                load_in_4bit=True,
+                                optimize_model=False,
+                                trust_remote_code=True)
+    logits_base_model = (model(input_ids)).logits
+
+    model = Model.from_pretrained(model_path,
+                                load_in_4bit=True,
+                                optimize_model=True,
+                                trust_remote_code=True)
+    logits_optimized_model = (model(input_ids)).logits
+    diff = abs(logits_base_model - logits_optimized_model).flatten()
+
+    assert any(diff) is False
+
+
+if __name__ == '__main__':
+    pytest.main([__file__])
diff --git a/python/llm/test/run-llm-inference-tests.sh b/python/llm/test/run-llm-inference-tests.sh
index 781bf44b..1c670e6b 100644
--- a/python/llm/test/run-llm-inference-tests.sh
+++ b/python/llm/test/run-llm-inference-tests.sh
@@ -9,13 +9,13 @@ set -e
 echo "# Start testing inference"
 start=$(date "+%s")
 
-python -m pytest -s ${LLM_INFERENCE_TEST_DIR} -k "not test_transformers"
+python -m pytest -s ${LLM_INFERENCE_TEST_DIR} -k "not test_transformers" -v
 
 if [ -z "$THREAD_NUM" ]; then
   THREAD_NUM=2
 fi
 export OMP_NUM_THREADS=$THREAD_NUM
-python -m pytest -s ${LLM_INFERENCE_TEST_DIR} -k test_transformers
+python -m pytest -s ${LLM_INFERENCE_TEST_DIR} -k test_transformers -v
 
 now=$(date "+%s")
 time=$((now-start))