o ^i% @sNddddddddddiZddd d Zd d d d ZddZddZdS)CUDAcK d|dS)Ng2CwidthkwargsrrP/var/www/html/RAG/RAG_venv/lib/python3.10/site-packages/triton/profiler/specs.py r cKr)NgbBrrrrrrr r cKs |d|ddd|dS)Nrgx:AigmBrrrnum_sms clock_raterrrrr scKs|d|dd|dS)Ni@@@g.Arrr rrrr s)808990100gvHBg xHBgB)gfx90agfx942gfx950iii cCsn|dkrt|||d|dS|tvrtd||t|vr*td|t||}||||dS)a Calculate the maximum FLOPS for a given device type and width. Args: device_type (str): The type of device (e.g., "CUDA", "HIP"). arch (str): The architecture of the device (e.g., "80", "90"). width (int): The width in bits. num_sms (int): The number of streaming multiprocessors. clock_rate (float): The clock rate in GHz. Returns: float: The maximum FLOPS for the given device type and width. HIPrrzUnsupported device type: zUnsupported architecture: )r r)amd_fp8_flops_by_archflops_by_device ValueError) device_typearchrr r flops_funcrrr max_flopss  rcCs0|dkrd||ddS|dksJt|S)a Calculate the maximum bytes per second for a given bus width and memory clock rate. Args: bus_width (int): The bus width in bits. memory_clock_rate (float): The memory clock rate in GHz. Returns: float: The maximum bytes per second. rrrr)amd_bps_by_arch)rr bus_widthmemory_clock_raterrrmax_bps6s  r#N)rr rrr#rrrrs