in src/optimum/nvidia/export/config.py [0:0]
def plugin_config(self) -> "PluginConfig":
config = PluginConfig()
config.gemm_plugin = "auto"
config.gpt_attention_plugin = "auto"
config.set_context_fmha(ContextFMHAType.enabled)
config.enable_paged_kv_cache(32)
config.use_paged_context_fmha = True
if self.sharding.world_size > 1:
config.set_nccl_plugin()
if DataType(self.dtype) == DataType.FLOAT8:
config.gemm_swiglu_plugin = True
return config