mirror of
https://github.com/NousResearch/atropos.git
synced 2026-04-19 12:57:58 +00:00
432 lines
14 KiB
Python
432 lines
14 KiB
Python
"""
|
|
PIQA (Physical Interaction QA) Evaluation Environment for Atropos
|
|
|
|
This environment evaluates models on the PIQA benchmark - testing
|
|
physical commonsense reasoning with binary choice questions.
|
|
|
|
Dataset: ybisk/piqa
|
|
Paper: https://arxiv.org/abs/1911.11641
|
|
|
|
PIQA tests:
|
|
- Physical commonsense reasoning
|
|
- Understanding how physical interactions work
|
|
- Binary choice (A or B) for completing a goal
|
|
- e.g., "How to boil water?" -> choose between two solutions
|
|
|
|
Metrics:
|
|
- Accuracy (exact match on A/B)
|
|
|
|
Supports optional thinking mode with <think></think> tags.
|
|
"""
|
|
|
|
import asyncio
|
|
from string import ascii_uppercase
|
|
from typing import Dict, List, Optional, Tuple
|
|
|
|
import wandb
|
|
from datasets import load_dataset
|
|
from eval_helpers import (
|
|
build_mcqa_fallback_patterns,
|
|
create_system_content,
|
|
extract_letter_from_answer_tag,
|
|
get_default_thinking_prompt,
|
|
save_eval_results,
|
|
validate_thinking_format,
|
|
)
|
|
from pydantic import Field
|
|
from tqdm.asyncio import tqdm_asyncio
|
|
|
|
from atroposlib.envs.base import (
|
|
APIServerConfig,
|
|
BaseEnv,
|
|
BaseEnvConfig,
|
|
)
|
|
|
|
|
|
class PIQAEvalConfig(BaseEnvConfig):
|
|
"""Configuration for PIQA evaluation environment."""
|
|
|
|
# Thinking mode configuration
|
|
thinking_mode: bool = Field(
|
|
default=True,
|
|
description="Whether to enable thinking mode with <think></think> tags.",
|
|
)
|
|
|
|
custom_thinking_prompt: Optional[str] = Field(
|
|
default=None,
|
|
description="Custom thinking prompt. If None, uses the default thinking prompt.",
|
|
)
|
|
|
|
# Dataset configuration
|
|
dataset_name: str = Field(
|
|
default="ybisk/piqa",
|
|
description="HuggingFace dataset name for PIQA.",
|
|
)
|
|
|
|
eval_split: str = Field(
|
|
default="validation",
|
|
description="Dataset split to use for evaluation (validation or test).",
|
|
)
|
|
|
|
# Model generation configuration
|
|
eval_temperature: float = Field(
|
|
default=0.6,
|
|
description="Temperature for model generation.",
|
|
)
|
|
|
|
eval_max_tokens: int = Field(
|
|
default=0,
|
|
description="Maximum tokens for evaluation responses. Set to 0 for provider default.",
|
|
)
|
|
|
|
# Prompt configuration
|
|
custom_system_prompt: Optional[str] = Field(
|
|
default=None,
|
|
description="Custom system prompt to append after thinking prompt.",
|
|
)
|
|
|
|
# Retry configuration
|
|
max_retries: int = Field(
|
|
default=3,
|
|
ge=1,
|
|
description="Maximum retries for failed API calls.",
|
|
)
|
|
|
|
retry_delay: float = Field(
|
|
default=1.0,
|
|
ge=0.0,
|
|
description="Delay between retry attempts in seconds.",
|
|
)
|
|
|
|
min_response_length: int = Field(
|
|
default=1,
|
|
ge=1,
|
|
description="Minimum response length to consider valid.",
|
|
)
|
|
|
|
# Debug configuration
|
|
full_debug: bool = Field(
|
|
default=False,
|
|
description="Enable verbose debug logging.",
|
|
)
|
|
|
|
|
|
class PIQAEvalEnv(BaseEnv):
|
|
"""
|
|
PIQA Evaluation Environment for Atropos.
|
|
|
|
Evaluates models on physical commonsense reasoning with binary choice.
|
|
"""
|
|
|
|
name = "piqa_eval"
|
|
env_config_cls = PIQAEvalConfig
|
|
|
|
def __init__(
|
|
self,
|
|
config: PIQAEvalConfig,
|
|
server_configs: List[APIServerConfig],
|
|
slurm=True,
|
|
testing=False,
|
|
):
|
|
super().__init__(config, server_configs, slurm, testing)
|
|
self.config: PIQAEvalConfig = config
|
|
self.eval_metrics = []
|
|
|
|
# Pre-build fallback patterns for 2-choice (A/B)
|
|
self._fallback_patterns = build_mcqa_fallback_patterns(2)
|
|
self._valid_letters = {"A", "B"}
|
|
|
|
@classmethod
|
|
def config_init(cls) -> Tuple[PIQAEvalConfig, List[APIServerConfig]]:
|
|
"""Initialize default configuration for CLI usage."""
|
|
config = PIQAEvalConfig(
|
|
tokenizer_name="NousResearch/Hermes-3-Llama-3.1-8B",
|
|
group_size=1,
|
|
use_wandb=True,
|
|
rollout_server_url="http://localhost:8000",
|
|
total_steps=1,
|
|
batch_size=1,
|
|
steps_per_eval=1,
|
|
max_token_length=2048,
|
|
wandb_name="piqa_eval",
|
|
data_path_to_save_groups=None,
|
|
eval_max_tokens=0,
|
|
)
|
|
server_config = APIServerConfig(
|
|
model_name="Hermes-3-Llama-3.1-8B",
|
|
base_url="http://localhost:8000/v1",
|
|
api_key="x",
|
|
num_requests_for_eval=1024,
|
|
)
|
|
return config, [server_config]
|
|
|
|
async def setup(self):
|
|
"""Load the PIQA dataset."""
|
|
print("\nPIQA Evaluation Setup (Generative Mode):")
|
|
print(f" Dataset: {self.config.dataset_name}")
|
|
print(f" Evaluation split: {self.config.eval_split}")
|
|
print(f" Thinking mode: {self.config.thinking_mode}")
|
|
if self.config.thinking_mode:
|
|
print(
|
|
f" Thinking prompt: {get_default_thinking_prompt(self.config.custom_thinking_prompt)[:80]}..."
|
|
)
|
|
|
|
# Load dataset
|
|
self.dataset = load_dataset(
|
|
self.config.dataset_name,
|
|
split=self.config.eval_split,
|
|
trust_remote_code=True,
|
|
)
|
|
|
|
self.eval_items = list(self.dataset)
|
|
print(f" Loaded {len(self.eval_items)} evaluation items")
|
|
|
|
def _format_prompt(self, item: Dict) -> Tuple[str, List[str]]:
|
|
"""
|
|
Format a PIQA item into a prompt.
|
|
|
|
PIQA has a goal and two solutions (sol1, sol2).
|
|
|
|
Returns the formatted prompt and list of choice texts.
|
|
"""
|
|
goal = item["goal"]
|
|
sol1 = item["sol1"]
|
|
sol2 = item["sol2"]
|
|
|
|
# Build the question
|
|
query = "The following is a multiple choice question about physical common sense.\n\n"
|
|
query += f"Goal: {goal}\n\n"
|
|
query += "Which solution is more likely to achieve the goal?\n"
|
|
query += f"A. {sol1}\n"
|
|
query += f"B. {sol2}\n"
|
|
query += "\nProvide your answer in <answer></answer> tags with only the letter (A or B)."
|
|
|
|
return query, [sol1, sol2]
|
|
|
|
def _create_system_content(self) -> Optional[str]:
|
|
"""Create system message content based on thinking mode configuration."""
|
|
return create_system_content(
|
|
self.config.thinking_mode,
|
|
self.config.custom_thinking_prompt,
|
|
self.config.custom_system_prompt,
|
|
)
|
|
|
|
def _extract_answer(
|
|
self, response: str, choices: List[str] = None
|
|
) -> Tuple[Optional[str], str]:
|
|
"""
|
|
Extract the answer letter from the model's response.
|
|
|
|
Uses <answer> tags as primary method, with fallback patterns.
|
|
"""
|
|
# Get content after </think> if in thinking mode
|
|
if self.config.thinking_mode:
|
|
is_valid, content_after_think = validate_thinking_format(response, True)
|
|
if is_valid:
|
|
response_to_parse = content_after_think
|
|
else:
|
|
response_to_parse = response
|
|
else:
|
|
response_to_parse = response
|
|
|
|
# Primary: Try <answer></answer> tags
|
|
letter, method = extract_letter_from_answer_tag(
|
|
response_to_parse,
|
|
self._valid_letters,
|
|
debug=self.config.full_debug,
|
|
choices=choices,
|
|
)
|
|
if letter:
|
|
return letter, method
|
|
|
|
# Fallback: Use regex patterns
|
|
for priority, pattern, method_name in self._fallback_patterns:
|
|
matches = pattern.findall(response_to_parse)
|
|
if matches:
|
|
match = (
|
|
matches[-1]
|
|
if method_name
|
|
in [
|
|
"final_answer_is",
|
|
"the_answer_is",
|
|
"answer_colon",
|
|
"answer_space",
|
|
]
|
|
else matches[0]
|
|
)
|
|
if isinstance(match, tuple):
|
|
match = match[0]
|
|
letter = match.strip("()").upper()
|
|
if letter in self._valid_letters:
|
|
return letter, f"fallback_{method_name}"
|
|
|
|
return None, "no_match"
|
|
|
|
async def _generate_with_retry(
|
|
self, messages: List[Dict], item_id: str
|
|
) -> Optional[str]:
|
|
"""Generate response with retry logic."""
|
|
for attempt in range(self.config.max_retries):
|
|
try:
|
|
api_params = {
|
|
"model": self.server_configs[0].model_name,
|
|
"messages": messages,
|
|
"temperature": self.config.eval_temperature,
|
|
}
|
|
if self.config.eval_max_tokens > 0:
|
|
api_params["max_tokens"] = self.config.eval_max_tokens
|
|
|
|
response = await self.client.chat.completions.create(**api_params)
|
|
|
|
if response.choices and response.choices[0].message.content:
|
|
content = response.choices[0].message.content.strip()
|
|
if len(content) >= self.config.min_response_length:
|
|
return content
|
|
|
|
except Exception as e:
|
|
if self.config.full_debug:
|
|
print(f" Error on item {item_id} attempt {attempt + 1}: {e}")
|
|
if attempt < self.config.max_retries - 1:
|
|
await asyncio.sleep(self.config.retry_delay * (attempt + 1))
|
|
|
|
return None
|
|
|
|
async def _evaluate_single_item(self, item: Dict, idx: int) -> Dict:
|
|
"""Evaluate a single PIQA item."""
|
|
# Format prompt
|
|
prompt, choices = self._format_prompt(item)
|
|
|
|
# Build messages
|
|
messages = []
|
|
system_content = self._create_system_content()
|
|
if system_content:
|
|
messages.append({"role": "system", "content": system_content})
|
|
messages.append({"role": "user", "content": prompt})
|
|
|
|
# Generate response
|
|
response = await self._generate_with_retry(messages, str(idx))
|
|
|
|
if response is None:
|
|
return {
|
|
"index": idx,
|
|
"is_correct": False,
|
|
"extracted_answer": None,
|
|
"gold_answer": (
|
|
ascii_uppercase[item["label"]] if item["label"] >= 0 else None
|
|
),
|
|
"extraction_method": "generation_failed",
|
|
"error": "Failed to generate response",
|
|
}
|
|
|
|
# Extract answer
|
|
extracted_answer, extraction_method = self._extract_answer(response, choices)
|
|
|
|
# Determine gold answer (PIQA uses 0/1 for label)
|
|
gold_answer = None
|
|
if item["label"] >= 0:
|
|
gold_answer = ascii_uppercase[item["label"]]
|
|
|
|
# Score
|
|
is_correct = (
|
|
extracted_answer == gold_answer
|
|
if extracted_answer and gold_answer
|
|
else False
|
|
)
|
|
|
|
result = {
|
|
"index": idx,
|
|
"is_correct": is_correct,
|
|
"extracted_answer": extracted_answer,
|
|
"gold_answer": gold_answer,
|
|
"extraction_method": extraction_method,
|
|
}
|
|
|
|
if self.config.full_debug:
|
|
result["response"] = response
|
|
result["prompt"] = prompt
|
|
|
|
return result
|
|
|
|
async def evaluate(self, *args, **kwargs):
|
|
"""Run the full PIQA evaluation."""
|
|
print("\n" + "=" * 60)
|
|
print("Starting PIQA Evaluation (Generative/Reasoning Mode)")
|
|
print("=" * 60)
|
|
print(f" Total questions: {len(self.eval_items)}")
|
|
print(f" Thinking mode: {self.config.thinking_mode}")
|
|
print("=" * 60)
|
|
|
|
# Evaluate all items
|
|
tasks = [
|
|
self._evaluate_single_item(item, idx)
|
|
for idx, item in enumerate(self.eval_items)
|
|
]
|
|
|
|
results = await tqdm_asyncio.gather(*tasks, desc="Evaluating PIQA")
|
|
|
|
# Calculate metrics
|
|
valid_results = [r for r in results if r.get("gold_answer") is not None]
|
|
|
|
if not valid_results:
|
|
print("Warning: No valid evaluation results obtained")
|
|
return
|
|
|
|
correct = sum(1 for r in valid_results if r["is_correct"])
|
|
total = len(valid_results)
|
|
accuracy = correct / total if total > 0 else 0.0
|
|
|
|
# Extraction method breakdown
|
|
method_counts = {}
|
|
for r in valid_results:
|
|
method = r.get("extraction_method", "unknown")
|
|
method_counts[method] = method_counts.get(method, 0) + 1
|
|
|
|
# Print summary
|
|
print("\n" + "=" * 60)
|
|
print("PIQA Evaluation Results")
|
|
print("=" * 60)
|
|
print(f" Total evaluated: {total}")
|
|
print(f" Correct: {correct}")
|
|
print(f" Accuracy: {accuracy:.2%}")
|
|
print("-" * 60)
|
|
print(" Extraction Methods:")
|
|
for method, count in sorted(method_counts.items(), key=lambda x: -x[1]):
|
|
print(f" {method}: {count} ({count/total:.1%})")
|
|
print("=" * 60)
|
|
|
|
# Save results
|
|
metrics = {
|
|
"accuracy": accuracy,
|
|
"total_evaluated": total,
|
|
"correct": correct,
|
|
"extraction_methods": method_counts,
|
|
}
|
|
|
|
save_eval_results(self.config.data_dir_to_save_evals, metrics, results)
|
|
|
|
self.eval_metrics = [
|
|
{
|
|
"accuracy": accuracy,
|
|
"total": total,
|
|
}
|
|
]
|
|
|
|
async def wandb_log(self, step: int):
|
|
"""Log metrics to wandb."""
|
|
if self.eval_metrics and wandb.run is not None:
|
|
for metric in self.eval_metrics:
|
|
wandb.log(metric, step=step)
|
|
|
|
# Required BaseEnv interface methods
|
|
async def get_next_item(self):
|
|
return None
|
|
|
|
async def collect_trajectories(self, *args, **kwargs):
|
|
return []
|
|
|
|
async def score(self, *args, **kwargs):
|
|
return []
|
|
|
|
|
|
if __name__ == "__main__":
|
|
PIQAEvalEnv.cli()
|