File size: 3,097 Bytes
bef2eac |
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 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 |
---
license: apache-2.0
language:
- en
---
This is a Re-act style model trained from meta-llama/Meta-Llama-3-8B
Dataset was parsed with:
```
def extract_trajectory_info(data):
"""
Extracts the question, thoughts, actions, and observations from the trajectory field of the data.
Parameters:
data (dict): The data entry containing the trajectory field.
Returns:
dict: A dictionary containing the extracted question, thoughts, actions, and observations.
"""
# Extracting the question
question = data.get('question', '')
# Extracting thoughts, actions, and observations using regex
thoughts = re.findall(r'Thought \d+: (.+?)(?=Action|\Z)', data.get('trajectory', ''), re.DOTALL)
actions = re.findall(r'Action \d+: (.+?)(?=Observation|\Z)', data.get('trajectory', ''), re.DOTALL)
observations = re.findall(r'Observation \d+: (.+?)(?=Thought|\Z)', data.get('trajectory', ''), re.DOTALL)
# Cleaning up the extracted data
thoughts = [thought.strip() for thought in thoughts]
actions = [action.strip() for action in actions]
observations = [observation.strip() for observation in observations]
return {
'question': question,
'thoughts': thoughts,
'actions': actions,
'observations': observations
}
# Sample data
extracted_info = extract_trajectory_info(ds["train"][0])
```
Then remade into a new dataset with
```
# Predefine the instructions for the task
preamble = """Tools available:
(1) Search[entity], which searches the exact entity on Wikipedia and returns the first paragraph if it exists. If not, it will return some similar entities to search.
(2) Lookup[keyword], which returns the next sentence containing the keyword in the current passage.
(3) Finish[answer], which returns the answer and finishes the task.
"""
dataset = []
# Iterate through a specified number of examples in the training set
for i in range(len(ds['train'])):
extracted_info = extract_trajectory_info(ds['train'][i])
# Iterate through each thought in the extracted information
for j in range(len(extracted_info['thoughts'])):
out = f"{preamble}---\nQuestion: {extracted_info['question']}\n"
prev = ""
# Construct output for the first thought
if j == 0:
out += f"Thought: {extracted_info['thoughts'][0]}\n"
out += f"Action: {extracted_info['actions'][0]}\nPAUSE\n\n\n\n"
else:
for k in range(1, j + 1):
# Use appropriate indexing to avoid out-of-bounds errors
prev += f"Thought:{extracted_info['thoughts'][j - k]}\n"
prev += f"Action: {extracted_info['actions'][j - k]}\nPAUSE\n"
prev += f"Observation: {extracted_info['observations'][j - k]}\n"
out += prev # Remove trailing space
out += f"---\nThought: {extracted_info['thoughts'][j]}\n"
out += f"Action: {extracted_info['actions'][j]}\nPAUSE\n\n\n\n"
# Print the constructed output
print(out)
dataset.append(out)
#print(len(out))
``` |