mirror of
https://github.com/huggingface/trl.git
synced 2025-10-21 02:53:59 +08:00
31b93876a7
* first piece of doc * improve readibility * some data utils and doc * simplify prompt-only * format * fix path data utils * fix example format * simplify * tests * prompt-completion * update antropic hh * update dataset script * implicit prompt * additional content * `maybe_reformat_dpo_to_kto` -> `unpair_preference_dataset` * Preference dataset with implicit prompt * unpair preference dataset tests * documentation * ... * doc * changes applied to dpo example * better doc and better log error * a bit more doc * improve doc * converting * some subsections * converting section * further refinements * tldr * tldr preference * rename * lm-human-preferences-sentiment * `imdb` to `stanfordnlp/imdb` * Add script for LM human preferences descriptiveness * Remove sentiment_descriptiveness.py script * style * example judge tlrd with new dataset * Syle * Dataset conversion for TRL compatibility * further refinements * trainers in doc * top level for functions * stanfordnlp/imdb * downgrade transformers * temp reduction of tests * next commit * next commit * additional content * proper tick format * precise the assistant start token * improve * lower case * Update titles in _toctree.yml and data_utils.mdx * revert make change * correct dataset ids * expand a bit dataset formats * skip gated repo tests * data utilities in API * Update docs/source/dataset_formats.mdx Co-authored-by: lewtun <lewis.c.tunstall@gmail.com> * Update docs/source/dataset_formats.mdx Co-authored-by: lewtun <lewis.c.tunstall@gmail.com> * Update docs/source/dataset_formats.mdx Co-authored-by: lewtun <lewis.c.tunstall@gmail.com> * Update docs/source/dataset_formats.mdx Co-authored-by: lewtun <lewis.c.tunstall@gmail.com> * tiny internal testing for chat template testing * precise type/format * exlude sft trainer in doc * Update trl/trainer/utils.py * XPO in the doc --------- Co-authored-by: lewtun <lewis.c.tunstall@gmail.com>
75 lines
2.7 KiB
Python
75 lines
2.7 KiB
Python
from dataclasses import dataclass, field
|
|
from typing import Optional
|
|
|
|
from datasets import load_dataset
|
|
from transformers import HfArgumentParser
|
|
from vllm import LLM, SamplingParams
|
|
|
|
from trl import HfPairwiseJudge, OpenAIPairwiseJudge
|
|
|
|
|
|
"""
|
|
Examples:
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/rloo_tldr --num_examples 1000
|
|
Model win rate: 31.40%
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/rloo_tldr --judge_model gpt-3.5-turbo-0125 --num_examples 1000
|
|
Model win rate: 51.60%
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/rloo_tldr --judge_model gpt-4o-mini --num_examples 1000
|
|
Model win rate: 51.20%
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/ppo_tldr --num_examples 1000
|
|
Model win rate: 46.30%
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/ppo_tldr --judge_model gpt-3.5-turbo-0125 --num_examples 1000
|
|
Model win rate: 52.50%
|
|
|
|
python examples/scripts/evals/judge_tldr.py --model_name_or_path vwxyzjn/ppo_tldr --judge_model gpt-4o-mini --num_examples 1000
|
|
Model win rate: 63.00%
|
|
"""
|
|
|
|
|
|
@dataclass
|
|
class ScriptArguments:
|
|
model_name_or_path: str = field(metadata={"help": "The model name or path to the model to evaluate."})
|
|
judge_model: str = field(
|
|
default="meta-llama/Meta-Llama-3-70B-Instruct",
|
|
metadata={
|
|
"help": "The model name or path to the model to use as a judge. E.g., 'gpt-3.5-turbo-0125', 'meta-llama/Meta-Llama-3-70B-Instruct'."
|
|
},
|
|
)
|
|
num_examples: Optional[int] = field(default=None, metadata={"help": "The number of examples to evaluate."})
|
|
|
|
|
|
# Parse the arguments
|
|
parser = HfArgumentParser(ScriptArguments)
|
|
args = parser.parse_args_into_dataclasses()[0]
|
|
|
|
# Load the dataset
|
|
raw_dataset = load_dataset("trl-lib/tldr", split="validation")
|
|
if args.num_examples is not None:
|
|
raw_dataset = raw_dataset.select(range(args.num_examples))
|
|
|
|
# Extract the prompts and reference completions
|
|
prompts = raw_dataset["prompt"]
|
|
reference_completions = raw_dataset["completion"]
|
|
|
|
# Generate the model completions
|
|
sampling_params = SamplingParams(temperature=0.0, top_p=0.95, max_tokens=200) # very generous max token length
|
|
llm = LLM(model=args.model_name_or_path, tensor_parallel_size=1)
|
|
outputs = llm.generate(prompts, sampling_params)
|
|
model_completions = [output.outputs[0].text.strip() for output in outputs]
|
|
|
|
# Judge the outputs
|
|
if "gpt" in args.judge_model:
|
|
judge = OpenAIPairwiseJudge(args.judge_model)
|
|
else:
|
|
judge = HfPairwiseJudge(args.judge_model)
|
|
|
|
completions = [[c0, c1] for c0, c1 in zip(reference_completions, model_completions)]
|
|
best_idxs = judge.judge(prompts, completions)
|
|
model_win_rate = best_idxs.count(1) / len(best_idxs)
|
|
print(f"Model win rate: {model_win_rate*100:.2f}%")
|