chore: import upstream snapshot with attribution

kt-kernel/python/sft/__init__.py

@@ -0,0 +1,83 @@
+# SFT (Supervised Fine-Tuning) submodule for kt-kernel
+# SPDX-License-Identifier: Apache-2.0
+
+"""
+SFT training support for KT-Kernel MoE.
+
+This submodule adds training capabilities (forward/backward, LoRA, autograd,
+distributed) on top of the inference-only kt_kernel base package.
+
+Additional dependencies beyond base kt_kernel: torch.nn, torch.distributed, peft (optional).
+"""
+
+from .config import KTConfig
+from .base import BaseSFTMoEWrapper, KExpertsSFTBuffer
+from .amx import AMXSFTMoEWrapper
+from .arch import (
+    MOEArchConfig, get_moe_arch_config, get_moe_module, move_non_experts_to_gpu, get_expert_device,
+    KTAMXError, KTAMXNotAvailableError, KTAMXModelNotSupportedError, KTAMXConfigError,
+)
+from .autograd import KTMoEFunction
+from .layer import KTMoELayerWrapper
+from .weights import (
+    extract_moe_weights,
+    load_experts_from_checkpoint_files,
+    load_experts_from_kt_weight_path,
+    INT8ExpertWeights,
+)
+from .lora import (
+    kt_adapt_peft_lora,
+    get_kt_lora_params,
+    update_kt_lora_pointers,
+    sync_kt_lora_gradients,
+    save_lora_experts_to_adapter,
+    save_kt_moe_to_adapter,
+    load_lora_experts_from_adapter,
+    load_kt_moe_from_adapter,
+    LoRAExpertMLP,
+    LoRAExperts,
+)
+from .wrapper import (
+    wrap_moe_layers_with_kt_wrapper,
+    build_kt_device_map,
+    build_kt_device_map_simplified,
+    get_kt_loading_kwargs,
+    load_kt_model,
+)
+
+__all__ = [
+    "KTConfig",
+    "BaseSFTMoEWrapper",
+    "KExpertsSFTBuffer",
+    "AMXSFTMoEWrapper",
+    "MOEArchConfig",
+    "get_moe_arch_config",
+    "get_moe_module",
+    "move_non_experts_to_gpu",
+    "get_expert_device",
+    "KTAMXError",
+    "KTAMXNotAvailableError",
+    "KTAMXModelNotSupportedError",
+    "KTAMXConfigError",
+    "KTMoEFunction",
+    "KTMoELayerWrapper",
+    "extract_moe_weights",
+    "load_experts_from_checkpoint_files",
+    "load_experts_from_kt_weight_path",
+    "INT8ExpertWeights",
+    "kt_adapt_peft_lora",
+    "get_kt_lora_params",
+    "update_kt_lora_pointers",
+    "sync_kt_lora_gradients",
+    "save_lora_experts_to_adapter",
+    "save_kt_moe_to_adapter",
+    "load_lora_experts_from_adapter",
+    "load_kt_moe_from_adapter",
+    "LoRAExpertMLP",
+    "LoRAExperts",
+    "wrap_moe_layers_with_kt_wrapper",
+    "build_kt_device_map",
+    "build_kt_device_map_simplified",
+    "get_kt_loading_kwargs",
+    "load_kt_model",
+]