1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
|
import json
from collaborativeagents.utils import get_conversation_string
from collaborativeagents.prompts import update_agent_notes_prompt
logiqa_llama70b_training_data = "/shared/storage-01/users/mehri2/mem/collaborativeagents/training/sft/training_data/unprocessed_conversations/logiqa_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_20.jsonl"
math_500_llama70b_training_data = "/shared/storage-01/users/mehri2/mem/collaborativeagents/training/sft/training_data/unprocessed_conversations/math_500_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_20.jsonl"
math_hard_llama70b_training_data = "/shared/storage-01/users/mehri2/mem/collaborativeagents/training/sft/training_data/unprocessed_conversations/math_hard_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_20.jsonl"
medqa_llama70b_training_data = "/shared/storage-01/users/mehri2/mem/collaborativeagents/training/sft/training_data/unprocessed_conversations/medqa_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_20.jsonl"
mmlu_llama70b_training_data = "/shared/storage-01/users/mehri2/mem/collaborativeagents/training/sft/training_data/unprocessed_conversations/mmlu_llama70b_user_llama70b_agent_training_data_with_reflection_eval_size_20.jsonl"
processed_data = []
for file_path in [logiqa_llama70b_training_data, math_500_llama70b_training_data, math_hard_llama70b_training_data, medqa_llama70b_training_data, mmlu_llama70b_training_data]:
unprocessed_data = []
with open(file_path, "r") as f:
for line in f:
unprocessed_data.append(json.loads(line))
for user_elem in unprocessed_data:
for conversation_elem in user_elem["generated_conversations"]:
conversation_str = get_conversation_string(conversation_elem["conversation"])
formatted_update_agent_notes_prompt = update_agent_notes_prompt.format(agent_notes="", conversation_str=conversation_str)
agent_notes_response = json.dumps(conversation_elem["agent_notes"], indent=2)
training_conversation = [
{"role": "user", "content": formatted_update_agent_notes_prompt},
{"role": "assistant", "content": agent_notes_response}
]
responses_that_enforce_preferences = [
elem["response"] for elem in conversation_elem["full_conversation_log"] if "enforce_preferences" in elem and elem["enforce_preferences"]
]
user_profile = {
"i": user_elem["i"],
"persona": user_elem["persona"],
"preferences": user_elem["preferences"]
}
processed_data.append({
"messages": training_conversation,
"responses_that_enforce_preferences": responses_that_enforce_preferences,
"user_profile": user_profile
})
with open("/shared/storage-01/users/mehri2/mem/collaborativeagents/training/grpo/training_data/session_level_reflection_grpo_data.jsonl", "w") as f:
json.dump(processed_data, f, indent=2)
|
