File size: 7,453 Bytes
5fed0fc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 |
"""
Model name utilities for Frontier-CS.
Provides consistent model prefix conversion used across:
- Solution generation (generate_solutions.py)
- Solution matrix checking (frontier-eval check)
- Batch evaluation (frontier-eval batch)
"""
import re
from typing import Dict, List, Optional, Tuple
# Model prefix aliases for backward compatibility
MODEL_PREFIX_ALIASES: Dict[str, str] = {
"grokcodefast1_": "grok4fastreasoning_",
}
def get_model_prefix(model: str) -> str:
"""
Convert model name to the prefix format used in solution folder names.
This is the canonical function for model prefix conversion. All tools
should use this to ensure consistent naming.
Examples:
>>> get_model_prefix("gpt-5")
'gpt5'
>>> get_model_prefix("gpt-5.1-preview")
'gpt5.1'
>>> get_model_prefix("gemini/gemini-2.5-pro")
'gemini2.5pro'
>>> get_model_prefix("claude-sonnet-4-5-20250929")
'claude4.5sonnet'
>>> get_model_prefix("grok-3-fast-reasoning")
'grok3fastreasoning'
Args:
model: Model name (e.g., "gpt-5", "gemini/gemini-2.5-pro")
Returns:
Normalized prefix for solution directory names
"""
original = model
# Remove provider prefix if present (e.g., 'gemini/gemini-2.5-pro' -> 'gemini-2.5-pro')
if "/" in model:
model = model.split("/", 1)[1]
model_lower = model.lower().strip()
# Handle GPT-5 variants
# Keep 'gpt-5.1' distinct so its artifacts prefix as 'gpt5.1'
if model_lower.startswith("gpt-5.1") or model_lower.startswith("gpt5.1"):
return "gpt5.1"
if model_lower.startswith("gpt-5") or model_lower.startswith("gpt5"):
return "gpt5"
# Handle Gemini 2.5 Pro variants
if "gemini-2.5-pro" in model_lower or "gemini2.5pro" in model_lower:
return "gemini2.5pro"
# Handle other Gemini variants (e.g., gemini-1.5-pro -> gemini1.5pro)
gemini_match = re.match(r"gemini-?(\d+\.?\d*)-?pro", model_lower)
if gemini_match:
version = gemini_match.group(1)
return f"gemini{version}pro"
# Handle Claude variants (e.g., claude-sonnet-4-5-20250929 -> claude4.5sonnet)
claude_match = re.match(r"claude-([a-z]+)-(\d+)-(\d+)", model_lower)
if claude_match:
family = claude_match.group(1)
major = claude_match.group(2)
minor = claude_match.group(3)
return f"claude{major}.{minor}{family}"
# Handle Grok variants - keep 'fast' and 'reasoning' in the prefix
if "grok" in model_lower:
sanitized = re.sub(r"[^a-zA-Z0-9]+", "", model_lower)
if sanitized:
return sanitized
# Default: sanitize by removing all non-alphanumeric characters
sanitized = re.sub(r"[^a-zA-Z0-9]+", "", model_lower)
if not sanitized:
raise ValueError(f"Unable to derive model prefix from '{original}'")
return sanitized
def normalize_solution_name(name: str) -> str:
"""
Normalize a solution directory name by applying prefix aliases.
Args:
name: Solution directory name
Returns:
Normalized name with aliases applied
"""
for old_prefix, new_prefix in MODEL_PREFIX_ALIASES.items():
if name.startswith(old_prefix):
return new_prefix + name[len(old_prefix):]
return name
def sanitize_problem_name(problem: str) -> str:
"""
Convert problem path to solution name suffix.
Examples:
>>> sanitize_problem_name("flash_attn")
'flash_attn'
>>> sanitize_problem_name("gemm_optimization/squares")
'gemm_optimization_squares'
Args:
problem: Problem ID (may contain slashes)
Returns:
Sanitized problem name for use in solution directory names
"""
return problem.replace("/", "_")
def parse_solution_name(solution_name: str) -> Tuple[str, str, int]:
"""
Parse a solution directory name into components.
Examples:
>>> parse_solution_name("gpt5_flash_attn")
('gpt5', 'flash_attn', 0)
>>> parse_solution_name("gpt5_flash_attn_1")
('gpt5', 'flash_attn', 1)
>>> parse_solution_name("claude4.5sonnet_gemm_optimization_squares_2")
('claude4.5sonnet', 'gemm_optimization_squares', 2)
Args:
solution_name: Solution directory name
Returns:
Tuple of (model_prefix, problem_slug, variant_index)
"""
# Check for variant suffix
parts = solution_name.rsplit("_", 1)
variant_index = 0
base_name = solution_name
if len(parts) == 2 and parts[1].isdigit():
variant_index = int(parts[1])
base_name = parts[0]
# Split into model prefix and problem slug
# Model prefix is the first part before underscore
first_underscore = base_name.find("_")
if first_underscore == -1:
return (base_name, "", variant_index)
model_prefix = base_name[:first_underscore]
problem_slug = base_name[first_underscore + 1:]
return (model_prefix, problem_slug, variant_index)
def build_solution_name(model: str, problem: str, variant_index: int = 0) -> str:
"""
Build a solution directory name from components.
Args:
model: Model name (will be converted to prefix)
problem: Problem ID
variant_index: Variant index (0 = no suffix)
Returns:
Solution directory name
"""
prefix = get_model_prefix(model)
slug = sanitize_problem_name(problem)
suffix = "" if variant_index == 0 else f"_{variant_index}"
return f"{prefix}_{slug}{suffix}"
def detect_provider(model: str) -> str:
"""
Detect the LLM provider from model name.
Args:
model: Model name
Returns:
Provider name: 'openai', 'google', 'anthropic', 'xai', 'deepseek', 'openrouter'
"""
normalized = model.strip()
if "/" in normalized:
provider_hint, actual_model = normalized.split("/", 1)
else:
provider_hint, actual_model = "", normalized
provider_hint = provider_hint.lower()
actual_lower = actual_model.lower()
if (provider_hint in {"", "openai", "azure", "azure_openai"}) and actual_lower.startswith("gpt"):
return "openai"
if provider_hint in {"gemini", "google"} or "gemini" in actual_lower:
return "google"
if provider_hint == "anthropic" or "claude" in actual_lower:
return "anthropic"
if provider_hint == "xai" or "grok" in actual_lower:
return "xai"
if provider_hint == "deepseek" or "deepseek" in actual_lower:
return "deepseek"
return provider_hint or "openai"
def is_reasoning_model(model: str, override: Optional[bool] = None) -> bool:
"""
Determine if a model is a reasoning model.
Args:
model: Model name
override: If set, use this value instead of auto-detection
Returns:
True if the model is a reasoning model
"""
if override is not None:
return override
prefixes = ("gpt-5", "o1", "o3", "deepseek-reasoner")
if any(model.startswith(p) for p in prefixes):
return True
normalized = model.lower()
if "reasoning" in normalized and normalized.startswith("grok-"):
return True
return False
|