Spaces:
Sleeping
Sleeping
File size: 1,711 Bytes
48a990e |
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 |
# llm_training_estimator.py
def get_gpu_teraflops(gpu_type: str, exo_flops: float = None) -> float:
"""
Returns TFLOPs/s based on GPU type or Exo input.
"""
gpu_lookup = {
"A100": 312, # FP16
"H100": 700,
}
if gpu_type == "Exo":
if exo_flops is None or exo_flops <= 0:
raise ValueError("Exo TFLOPs must be provided.")
return exo_flops
return gpu_lookup.get(gpu_type, 0)
def get_gpu_cost_per_tflop_hour(gpu_type: str, manual_cost: float = None) -> float:
"""
Returns ₹ cost per TFLOP-hour based on GPU type.
"""
cost_lookup = {
"A100": 17.64 * 83, # $17.64 × ₹83 (approx.) = ₹1464 per TFLOP/hr
"H100": 7.56 * 83 # ₹627 per TFLOP/hr
}
if gpu_type == "Exo" and manual_cost:
return manual_cost
return cost_lookup.get(gpu_type, 0)
def estimate_training_time_and_cost(params_billion: float, tokens_billion: float, teraflops: float,
cost_per_tflop_hr: float = 0.0):
"""
Estimate training time and cost.
Returns:
- FLOPs required
- Time (hours/days)
- Cost (₹)
"""
total_flops_required = 6 * params_billion * 1e9 * tokens_billion * 1e9 # 6 × N × T
time_seconds = total_flops_required / (teraflops * 1e12)
time_hours = time_seconds / 3600
time_days = time_hours / 24
total_cost = (teraflops * time_hours) * cost_per_tflop_hr if cost_per_tflop_hr > 0 else 0
return {
"flops_required": total_flops_required,
"time_hours": time_hours,
"time_days": time_days,
"total_cost": total_cost
}
|