refactor: move all runnable script to codecritic.cli

codecritic/cli/mk_cov_dataset.py

+import argparse
+from itertools import chain
+
+from codecritic.data.cov import (
+    convert_preference_to_vot_prompt,
+    convert_cov_to_cov_dataset,
+)
+
+from codecritic.utils.json import load_json
+from codecritic.utils.data import save_jsonl_dataset
+from codecritic.utils.vllm import vllm_chatcomplete
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model", type=str)
+    parser.add_argument("--preference_dataset", type=str)
+    parser.add_argument("--output_dir", type=str)
+    args = parser.parse_args()
+
+    preference_dataset = load_json(args.preference_dataset)
+    cov_prompts = [convert_preference_to_vot_prompt(x) for x in preference_dataset]
+    cov_prompts = list(chain(*cov_prompts))
+
+    sampling_params = dict(n=1, temperature=0.0, max_tokens=2048)
+    covs = vllm_chatcomplete(args.model, cov_prompts, sampling_params)
+    dataset = list(map(convert_cov_to_cov_dataset, covs))
+
+    save_jsonl_dataset(dataset, args.output_dir)

codecritic/data/cov.py

 # Additional Experiment:
 # Is reasoning really work? Let's verify step by step.
-import argparse
-from itertools import chain
 
-from codecritic.utils.json import load_json
 from codecritic.utils.data import (
     extract_code,
     code_template,
     mk_message,
     mk_messages,
     mk_critic_verify,
-    save_jsonl_dataset,
     SPLITTER,
 )
 
@@ -80,23 +76,3 @@ def convert_cov_to_cov_dataset(item):
         raise ValueError("Invalid prompt")
     item["messages"] += mk_critic_verify(is_correct)
     return item
-
-
-if __name__ == "__main__":
-    from codecritic.utils.vllm import vllm_chatcomplete
-
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--model", type=str)
-    parser.add_argument("--preference_dataset", type=str)
-    parser.add_argument("--output_dir", type=str)
-    args = parser.parse_args()
-
-    preference_dataset = load_json(args.preference_dataset)
-    cov_prompts = [convert_preference_to_vot_prompt(x) for x in preference_dataset]
-    cov_prompts = list(chain(*cov_prompts))
-
-    sampling_params = dict(n=1, temperature=0.0, max_tokens=2048)
-    covs = vllm_chatcomplete(args.model, cov_prompts, sampling_params)
-    dataset = list(map(convert_cov_to_cov_dataset, covs))
-
-    save_jsonl_dataset(dataset, args.output_dir)

codecritic/utils/data.py

 import re
-from codecritic.utils.json import save_jsonl
 from pathlib import Path
 
+from codecritic.utils.json import save_jsonl
+
 codeblock_pattern = re.compile(r"```python(.+?)```", flags=re.DOTALL)
 code_template = """```python
 {}
@@ -69,22 +70,4 @@ def get_score_token_id(tokenizer, token_str="Yes"):
     return score_tokens[0]
 
 
-def mk_critic_reason(codedit, explanation):
-    user_question = {
-        "role": "user",
-        "content": "Edit your code in diff format to fix any issues and explain the changes.",
-    }
-    llm_answer_content = f"""\
-**Edited Code (in diff format):**
-```diff
-{codedit}
-```
-
-**Explanation:**
-{explanation}
-"""
-    llm_answer = {"role": "assistant", "content": llm_answer_content}
-    return [user_question, llm_answer]
-
-
 SPLITTER = "__I_wish_it_were_weekends_all_the_time.__"