Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Quick fix on GPT4-eval #1696

Merged
merged 2 commits into from
Jun 5, 2024
Merged

Quick fix on GPT4-eval #1696

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
08b116d
quick fix
vwxyzjn Jun 5, 2024
e94ff3b
precommit
vwxyzjn Jun 5, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 10 additions & 4 deletions examples/scripts/evals/generate_tldr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
from datasets import load_dataset
from gpt_tldr_judge import LLMJudgeConfig, llm_judge
from transformers import AutoTokenizer, HfArgumentParser
from vllm import SamplingParams, SingleGPULLM
from vllm import LLM, SamplingParams


"""
Expand All @@ -28,6 +28,7 @@ class Args:
output_path: str
model_name_or_path: str
model_revision: str = "main"
judge_model: str = "gpt-3.5-turbo-0125"
n: int = 1000


Expand All @@ -50,11 +51,11 @@ def run_command(command: str):
prompts = prompts[: args.n]
reference_summaries = [message[-1]["content"] for message in raw_datasets["test"]["messages"]]
sampling_params = SamplingParams(temperature=0.0, top_p=0.95, max_tokens=MAX_TOKENS)
llm = SingleGPULLM(
llm = LLM(
model=args.model_name_or_path,
revision=args.model_revision,
tokenizer_revision=args.model_revision,
tensor_parallel_size=1,
device="cuda:0",
)
outputs = llm.generate(prompts, sampling_params)
table = defaultdict(list)
Expand Down Expand Up @@ -82,8 +83,13 @@ def run_command(command: str):
judged_df = llm_judge(
LLMJudgeConfig(
n=args.n,
model="gpt-3.5-turbo-0125",
model=args.judge_model,
),
df,
)
judged_df.rename(columns={"response0": "model_response", "response1": "reference_response"}, inplace=True)
print(judged_df["preferred"].value_counts())
# print percentage
print(judged_df["preferred"].value_counts(normalize=True))

judged_df.to_csv(args.output_path.replace(".csv", "_judged.csv"))
2 changes: 1 addition & 1 deletion examples/scripts/evals/gpt_tldr_judge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,7 +124,6 @@ async def main(ljc: LLMJudgeConfig, df: pd.DataFrame):
else "response1"
)
df.at[i, "preferred"] = preferred_label
print(df["preferred"].value_counts())
return df

return asyncio.run(main(ljc, df))
Expand All @@ -138,4 +137,5 @@ async def main(ljc: LLMJudgeConfig, df: pd.DataFrame):
df["response0"] = df["model_response"].map(lambda x: x.strip())
df["response1"] = df["reference_response"].map(lambda x: x.strip())
judge_df = llm_judge(ljc, df)
print(judge_df["preferred"].value_counts())
judge_df.to_csv(args.output_path)
Loading