| from model_inference import * |
| from config import dataset_files, cot_files |
| from result_parser import find_option_number_for_cot |
| from tqdm import tqdm |
| import json |
| import os |
|
|
| models = [gemma2b] |
| tasks = ["urban_region_function_recognition", "trajectory_region", "trajectory_trajectory", "trajectory_classification"] |
|
|
| if not os.path.exists("./logs"): |
| os.mkdir("./logs") |
|
|
| for fun in models: |
| model = fun() |
| for task in tasks: |
| error_writer = open("./logs/cot_{}.log".format(task), 'a') |
| error_writer.write(model.model_path+'\n') |
|
|
| context_samples = open(cot_files[task]) |
| prompt = "" |
| for _i, sample in enumerate(context_samples.readlines()): |
| sample = json.loads(sample) |
| prompt += "{}{}\n".format(sample['Question'], sample['Answer']) |
|
|
| for dataset_path in dataset_files[task]: |
| dataset = open(dataset_path, 'r') |
| dataset = dataset.readlines() |
|
|
| correct = 0 |
| total = 0 |
| exception = 0 |
|
|
| for i, item in tqdm(enumerate(dataset), total=len(dataset)): |
| item = json.loads(item) |
|
|
| |
| if task=="urban_region_function_recognition": |
| question = item['Question'].replace("Please just answer the number of your option with no other texts. Answer: Option (", "") |
| elif task=="trajectory_trajectory": |
| question = item['Question'].replace(" with no other texts. Answer: Option (", ".") |
| elif task=="trajectory_region": |
| question = item['Question'].replace(" with no other texts. Answer: Option ", ".") |
| elif task=="trajectory_classification": |
| question = item['Question'].replace("Answer: The trajectory is most likely to be generated by", "") |
|
|
| response = model.generate(prompt+question, 100) |
| score = find_option_number_for_cot(response, item["Answer"], error_writer) |
| |
| total +=1 |
| if score is None: |
| exception += 1 |
| else: |
| correct += score |
|
|
| if i%100==0: |
| print("Dataset: {}\nTotal: {}, correct:{}, exception:{}, accuracy:{}\n\n".format(dataset_path, total, correct, exception, correct/total)) |
| |
| error_writer.write("Dataset: {}\nTotal: {}, correct:{}, exception:{}, accuracy:{}\n\n".format(dataset_path, total, correct, exception, correct/total)) |
| error_writer.flush() |
| error_writer.write("\n") |
| error_writer.close() |
|
|
