Add Llama2 Onnx Model E2E test

onnxruntime/python/tools/transformers/models/llama/test/infer_onnx_model.py

@@ -0,0 +1,45 @@
+import argparse
+
+from optimum.onnxruntime import ORTModelForCausalLM
+from transformers import AutoTokenizer, pipeline
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model_dir",
+        required=True,
+        help="model direcotory, including config.json and tokenizer.model",
+    )
+    parser.add_argument(
+        "--prompt",
+        required=True,
+        help="prompt string for the model to generate text from. e.g. 'question: What is the lightest element?'",
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+    model_dir = args.model_dir
+
+    model = ORTModelForCausalLM.from_pretrained(model_dir)
+    tokenizer = AutoTokenizer.from_pretrained(model_dir)
+    ort_llama2_generator = pipeline("text-generation", model=model, tokenizer=tokenizer, device="cuda:0")
+
+    sequences = ort_llama2_generator(
+        args.prompt,
+        do_sample=False,
+        num_return_sequences=1,
+        eos_token_id=tokenizer.eos_token_id,
+        pad_token_id=tokenizer.pad_token_id,
+        max_new_tokens=256,
+        return_full_text=False,
+        repetition_penalty=1.1,
+    )
+    print(sequences[0]["generated_text"])
+
+
+if __name__ == "__main__":
+    main()

onnxruntime/python/tools/transformers/models/llama/test/llama2-7b-fp16/config.json

@@ -0,0 +1,28 @@
+{
+  "_name_or_path": "/home/azureuser/git/onnxruntime/onnxruntime/python/tools/transformers/Llama-2-7b-hf/config.json",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 32,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "float16",
+  "transformers_version": "4.37.2",
+  "use_cache": true,
+  "vocab_size": 32000
+}

onnxruntime/python/tools/transformers/models/llama/test/llama2-7b-fp16/expected_answer.txt

@@ -0,0 +1,6 @@
+
+The answer to this question is that hydrogen is the lightest element. Hydrogen has an atomic number of 1 and a mass number of 1, which means it only contains one proton in its nucleus. This makes it much lighter than other elements with higher atomic numbers or mass numbers.
+Hydrogen also has a very low density compared to other elements due to its small size and low atomic weight (one-twelfth that of carbon). As such, it can be found in many different forms including gas, liquid, solid crystals and even plasma formations within stars like our sun!
+What Is The Lightest Element In The Periodic Table?
+The lightest element in the periodic table is helium. Helium is a noble gas and has two electrons in its outer shell, making it unreactive with most substances. It’s also extremely rare on Earth because it’s so lightweight – just one atom weighs less than four hydrogen atoms combined!
+Helium was first discovered by Sir William Ramsay and Lord Rayleigh in 1895 when they isolated samples from air using liquefaction techniques at low temperatures (-269°C).

onnxruntime/python/tools/transformers/models/llama/test/llama2-7b-fp16/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}