File size: 6,201 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 |
"""Environment configuration for solution generation."""
from dataclasses import dataclass
from pathlib import Path
from typing import Dict, Optional
from frontier_cs.config import load_runtime_config, get_effective_gpu_type
DEFAULT_GPU_TYPE = "L4"
# GPU specifications mapping (SkyPilot-compatible GPU types)
GPU_SPECS: Dict[str, Dict[str, str]] = {
"L4": {"name": "NVIDIA L4", "vram": "24GB"},
"A10G": {"name": "NVIDIA A10G", "vram": "24GB"},
"A100": {"name": "NVIDIA A100", "vram": "40GB"},
"A100-40GB": {"name": "NVIDIA A100", "vram": "40GB"},
"A100-80GB": {"name": "NVIDIA A100", "vram": "80GB"},
"H100": {"name": "NVIDIA H100", "vram": "80GB"},
"V100": {"name": "NVIDIA V100", "vram": "16GB"},
"V100-32GB": {"name": "NVIDIA V100", "vram": "32GB"},
"T4": {"name": "NVIDIA T4", "vram": "16GB"},
}
# Base system prompt template - environment section will be injected
SYSTEM_PROMPT_TEMPLATE = """You are an expert programmer. Generate Python code for the given problem.
{environment_section}
REQUIREMENTS:
1. Output ONLY Python code - no explanations, no markdown
2. Implement ALL required classes/functions from the API section
3. Use efficient algorithms appropriate for the evaluation environment
4. Final class name must match the API specification exactly
Output ONLY the code, starting with imports."""
@dataclass
class EnvConfig:
"""Environment configuration."""
gpu_type: Optional[str] = None # None = CPU, string = GPU type
environment: Optional[str] = None # Problem-specific environment description
# Resources from config.yaml
cpus: Optional[str] = None # e.g., "8", "8+", "4-8"
memory: Optional[str] = None # e.g., "32", "32+"
disk_size: Optional[int] = None # GB
instance_type: Optional[str] = None # e.g., "n1-standard-8"
def _format_spec(spec: str) -> str:
"""Format a spec like '8+' to '8+ (or more)'."""
if spec.endswith("+"):
return f"{spec[:-1]}+ (or more)"
return spec
def build_cpu_environment(config: EnvConfig) -> str:
"""Generate CPU environment description."""
cpu_spec = config.cpus or "8"
mem_spec = config.memory or "16"
cpu_display = _format_spec(cpu_spec)
mem_display = _format_spec(mem_spec)
lines = ["EVALUATION ENVIRONMENT:"]
if config.instance_type:
lines.append(f"- Instance: {config.instance_type}")
lines.append(f"- CPU-only: {cpu_display} vCPUs, {mem_display}GB RAM (NO GPU)")
if config.disk_size:
lines.append(f"- Disk: {config.disk_size}GB")
if config.environment:
lines.append(f"- {config.environment}")
return "\n".join(lines)
def build_gpu_environment(config: EnvConfig) -> str:
"""Generate GPU environment description from config."""
gpu_type = config.gpu_type or DEFAULT_GPU_TYPE
spec = GPU_SPECS.get(gpu_type, GPU_SPECS[DEFAULT_GPU_TYPE])
lines = ["EVALUATION ENVIRONMENT:"]
if config.instance_type:
lines.append(f"- Instance: {config.instance_type}")
lines.append(f"- GPU: {spec['name']} ({spec['vram']} VRAM)")
if config.cpus or config.memory:
cpu_spec = config.cpus or "8"
mem_spec = config.memory or "32"
cpu_display = _format_spec(cpu_spec)
mem_display = _format_spec(mem_spec)
lines.append(f"- CPU: {cpu_display} vCPUs, {mem_display}GB RAM")
if config.disk_size:
lines.append(f"- Disk: {config.disk_size}GB")
if config.environment:
lines.append(f"- {config.environment}")
return "\n".join(lines)
def load_env_config_from_problem(problem_path: Path) -> EnvConfig:
"""
Load environment configuration from problem's config.yaml runtime section.
Supported config.yaml runtime fields:
- gpu_type: SkyPilot GPU type (e.g., "L4", "A100")
- resources.accelerators: SkyPilot accelerators (e.g., "L4:1", "A100:4")
- resources.cpus: CPU specification (e.g., "8", "8+")
- resources.memory: Memory in GB (e.g., "32", "32+")
- resources.disk_size: Disk size in GB
- resources.instance_type: Cloud instance type
- environment: Problem-specific environment description
"""
env_config = EnvConfig()
runtime_config = load_runtime_config(problem_path)
gpu_type = get_effective_gpu_type(runtime_config)
if gpu_type:
env_config.gpu_type = gpu_type
res = runtime_config.resources
if res.cpus:
env_config.cpus = res.cpus
if res.memory:
env_config.memory = res.memory
if res.disk_size:
env_config.disk_size = res.disk_size
if res.instance_type:
env_config.instance_type = res.instance_type
if runtime_config.environment:
env_config.environment = runtime_config.environment
return env_config
def get_system_prompt_for_problem(
problem_name: str,
problem_path: Optional[Path] = None,
docker_config: Optional[Dict] = None,
) -> str:
"""
Build system prompt with environment info.
Priority (with fallbacks):
1. config.yaml runtime section -> Use specified values
2. docker_images.txt GPU detection -> GPU with default config, or CPU
3. Default CPU environment
"""
env_config = EnvConfig()
# Priority 1: Try to load config from config.yaml
if problem_path and problem_path.is_dir():
env_config = load_env_config_from_problem(problem_path)
# Priority 2: Fallback to docker_images.txt for GPU detection
if env_config.gpu_type is None and docker_config:
base_name = problem_name.split("/")[0] if "/" in problem_name else problem_name
if "_" in base_name and base_name not in docker_config:
base_name = base_name.split("_")[0]
if base_name in docker_config:
_, gpu_enabled, _ = docker_config[base_name]
if gpu_enabled:
env_config.gpu_type = DEFAULT_GPU_TYPE
# Build environment section based on GPU or CPU mode
if env_config.gpu_type:
environment_section = build_gpu_environment(env_config)
else:
environment_section = build_cpu_environment(env_config)
return SYSTEM_PROMPT_TEMPLATE.format(environment_section=environment_section)
|