Created
June 16, 2024 16:12
-
-
Save ftnext/ac42ed33019cf4856a7a3437f91a2dfc to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# /// script | |
# dependencies = ["openai", "tqdm"] | |
# /// | |
# based on https://github.com/rasbt/LLMs-from-scratch/blob/aba7ed2eb1fce4ebbca28eeed11ab19687cb1764/ch07/03_model-evaluation/llm-instruction-eval-ollama.ipynb | |
from openai import OpenAI | |
from tqdm import tqdm | |
json_data = [ | |
{ | |
"instruction": "Name 3 different animals that are active during the day.", | |
"input": "", | |
"output": "1. Squirrel\n2. Eagle\n3. Tiger", | |
"model 1 response": "\n1. Squirrel\n2. Tiger\n3. Eagle\n4. Cobra\n5. Tiger\n6. Cobra", | |
"model 2 response": "\n1. Squirrel\n2. Eagle\n3. Tiger", | |
}, | |
{ | |
"instruction": "Take the following noun and turn it into a verb.", | |
"input": "Light", | |
"output": "To light.", | |
"model 1 response": "\nTo light.", | |
"model 2 response": "\nLight is the most abundant element in the periodic table.", | |
}, | |
] | |
def format_input(entry): | |
instruction_text = f"""\ | |
Below is an instruction that describes a task. Write a response that appropriately completes the request. | |
### Instruction: | |
{entry["instruction"]}\ | |
""" | |
input_text = f"\n\n### Input:\n{entry['input']}" if entry["input"] else "" | |
return instruction_text + input_text | |
client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama") | |
for entry in tqdm(json_data, desc="Scoring entries"): | |
input_string = format_input(entry) | |
print(input_string) | |
print() | |
print(f"correct output: {entry['output']}") | |
print() | |
for json_key in ("model 1 response", "model 2 response"): | |
prompt = f"Given the input `{input_string}`, and correct output `{entry['output']}`, score the model response `{entry[json_key]}` on a scale from 0 to 100, where 100 is the best score. Respond with the integer number only." | |
response = client.chat.completions.create( | |
model="llama3", | |
seed=123, | |
temperature=0, | |
max_tokens=5, | |
messages=[ | |
{"role": "user", "content": prompt}, | |
], | |
) | |
try: | |
score = int(response.choices[0].message.content) | |
except ValueError: | |
continue | |
print(f"{json_key}: {entry[json_key]}") | |
print(f"score {score}/100") | |
print() | |
print("-" * 40) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment