andylizf's picture
Upload folder using huggingface_hub
5fed0fc verified
"""
Pair expansion and management for batch evaluation.
A Pair represents a (solution, problem) combination to evaluate.
"""
import hashlib
from dataclasses import dataclass
from pathlib import Path
from typing import List, Optional
from ..models import get_model_prefix, sanitize_problem_name
@dataclass
class Pair:
"""Represents a solution-problem pair for evaluation."""
solution: str # Solution identifier (e.g., "gpt5_flash_attn")
problem: str # Problem identifier (e.g., "flash_attn")
@property
def id(self) -> str:
"""Unique identifier for this pair."""
return f"{self.solution}:{self.problem}"
@property
def safe_name(self) -> str:
"""Filesystem-safe name for this pair."""
base = f"{self.solution}-{self.problem}"
digest = hashlib.md5(base.encode("utf-8")).hexdigest()[:8]
sanitized = _sanitize_name(base)
suffix = f"-{digest}"
max_len = 63
available = max_len - len(suffix)
trimmed = sanitized[:available].rstrip("-")
return _sanitize_name(f"{trimmed}{suffix}")
def __hash__(self):
return hash(self.id)
def __eq__(self, other):
if isinstance(other, Pair):
return self.id == other.id
return False
def _sanitize_name(name: str) -> str:
"""Sanitize a name to be a valid cluster/file name."""
cleaned = []
valid = "abcdefghijklmnopqrstuvwxyz0123456789-"
last_dash = False
for ch in name.lower():
if ch in valid:
cleaned.append(ch)
last_dash = ch == "-"
else:
if not last_dash:
cleaned.append("-")
last_dash = True
sanitized = "".join(cleaned).strip("-")
return sanitized or "job"
# get_model_prefix and sanitize_problem_name imported from ..models
def expand_pairs(
problems: List[str],
models: List[str],
variants: Optional[List[int]] = None,
*,
solutions_dir: Optional[Path] = None,
validate_paths: bool = True,
) -> List[Pair]:
"""
Expand problems × models × variants into pairs.
Args:
problems: List of problem IDs (e.g., ["flash_attn", "cross_entropy"])
models: List of model names (e.g., ["gpt-5", "claude-sonnet-4-5"])
variants: List of variant indices (default: [0] for no suffix)
solutions_dir: Directory containing solutions (for validation)
validate_paths: Whether to validate solution paths exist
Returns:
List of Pair objects
"""
if variants is None:
variants = [0]
pairs: List[Pair] = []
for problem in problems:
problem_name = sanitize_problem_name(problem)
for model in models:
model_prefix = get_model_prefix(model)
for variant_idx in variants:
suffix = "" if variant_idx == 0 else f"_{variant_idx}"
solution_name = f"{model_prefix}_{problem_name}{suffix}"
if validate_paths and solutions_dir:
solution_path = solutions_dir / solution_name
if not solution_path.exists():
continue
pairs.append(Pair(solution=solution_name, problem=problem))
return pairs
def read_pairs_file(path: Path) -> List[Pair]:
"""
Read pairs from a pairs file.
Format: one pair per line as "solution:problem"
Lines starting with # are comments.
"""
pairs: List[Pair] = []
with path.open("r", encoding="utf-8") as f:
for line in f:
stripped = line.strip()
if not stripped or stripped.startswith("#"):
continue
if ":" not in stripped:
raise ValueError(f"Invalid pair line (expected solution:problem): {stripped}")
solution, problem = stripped.split(":", 1)
pairs.append(Pair(solution=solution.strip(), problem=problem.strip()))
return pairs
def read_problems_file(path: Path) -> List[str]:
"""Read problems from a problems file (one per line)."""
problems: List[str] = []
with path.open("r", encoding="utf-8") as f:
for line in f:
stripped = line.strip()
if not stripped or stripped.startswith("#"):
continue
# Normalize: remove 'research/' prefix if present
if stripped.startswith("research/"):
stripped = stripped[len("research/"):]
problems.append(stripped)
return problems
def read_models_file(path: Path) -> List[str]:
"""Read models from a models file (one per line)."""
models: List[str] = []
with path.open("r", encoding="utf-8") as f:
for line in f:
stripped = line.strip()
if not stripped or stripped.startswith("#"):
continue
models.append(stripped)
return models
def read_variants_file(path: Path) -> List[int]:
"""Read variant indices from a file (one per line)."""
variants: List[int] = []
if not path.exists():
return [0] # Default: just index 0 (no suffix)
with path.open("r", encoding="utf-8") as f:
for line in f:
stripped = line.strip()
if not stripped or stripped.startswith("#"):
continue
try:
variants.append(int(stripped))
except ValueError:
pass
return variants if variants else [0]
def read_solution_config(solution_dir: Path) -> Optional[str]:
"""
Read problem from a solution's config.yaml.
Returns:
Problem path (e.g., "flash_attn") or None if not found.
"""
config_file = solution_dir / "config.yaml"
if not config_file.exists():
return None
try:
import yaml
with config_file.open("r", encoding="utf-8") as f:
config = yaml.safe_load(f)
if config and isinstance(config, dict):
return config.get("problem")
except Exception:
# Fallback: simple parsing for "problem: value"
try:
content = config_file.read_text(encoding="utf-8")
for line in content.splitlines():
line = line.strip()
if line.startswith("problem:"):
return line.split(":", 1)[1].strip()
except Exception:
pass
return None
def scan_solutions_dir(solutions_dir: Path) -> List[Pair]:
"""
Scan solutions directory and build pairs from config.yaml files.
Args:
solutions_dir: Path to solutions directory
Returns:
List of Pair objects for solutions that have valid config.yaml
"""
pairs: List[Pair] = []
if not solutions_dir.is_dir():
return pairs
for solution_path in sorted(solutions_dir.iterdir()):
if not solution_path.is_dir() or solution_path.name.startswith("."):
continue
problem = read_solution_config(solution_path)
if problem:
pairs.append(Pair(solution=solution_path.name, problem=problem))
return pairs