LightCode

This assumes you have already done the Installation Guide.

Switch to the schedule conda envuroemnt

Raw sim results

Complete code can be found ()

Import lightcode

from lightcode import main
from lightcode import models

Define youre hardware

CPU_AVERAGE_CLOCK = 3.208 * 10**9  # 60**9, 6
PHU_MIN_CLOCK = 9.7 * 10**9  # 100**9, 10 Ghz
GPU_FP32_CLOCK = 1.98 * 10**9  # 1.98 GHz

local_hardware = []
main.hardware.Hardware._hardware_reset()
local_hardware.append(main.hardware.CPU(CPU_AVERAGE_CLOCK, 1)) # 1 core
local_hardware.append(main.hardware.PHU(PHU_MIN_CLOCK, 1, 20)) # 1 core, 20 multiplex factor

GPC = 8  # Graphical Processing Clusters
TPC_per_GPC = 9  # Texture Processing Clusters/Graphical Processing Cluster
SM_per_TPC = 2  # Streaming multiprocessors / Texture Processing Cluster
fp32_CUDA_cores_per_SM = 128  # fp32_CUDA_cores / Streaming multiprocessor
TC_per_SM = 4  # Tensor Cores / Streaming multiprocessor
local_hardware.append(main.hardware.GPU(
        GPU_FP32_CLOCK,
        GPC,
        TPC_per_GPC,
        SM_per_TPC,
        fp32_CUDA_cores_per_SM,
        TC_per_SM,
    )
)

available_hardware = main.hardware.initilize_hardware(local_hardware)

• NOTE: Definig a hardware type (hardwar.CPU, hardware.PHU, etc.) adds that hardware to the sims resources. The sim will expect that hardware to be part of the initilization. Before anohter sim with different hardware paramatrs, you need to run the hardware.Hardware._hardware_reset() and redefine and reinitilize the new hardware.

Define optimization paramater and run graph search

optimization = "time"
# optimization = "energy"
# optimization = "always_phu"

ans = main.graph_search(
    models.gpt2_prefill,
    optimization,
    available_hardware,
    moc_sequence_length=50,
    profiles=True,
    colect_data=True,
)

{‘moc_sequence_length’: 50, ‘Makespan’: 0.04984, ‘num_nodes’: 1518, ‘total_energy’: 43079811496.0, ‘num_photonic’: 73, ‘posiable_photonic’: 73}

Plots