[2026春季][T2-2-1] surprisely

[surprisely]

Summary

本 PR 为 InfiniTrain 增加统一的 Generator / GeneratorImpl 随机数基础设施，补齐 CPU、CUDA、逐设备默认 Generator、全局 seed、state 保存恢复，以及显式/默认 Generator 两种调用路径。

随机能力已接入初始化和框架随机张量入口：

• Tensor::Uniform；
• nn::init::{Uniform, Normal, KaimingUniform}；
• nn::function::{Rand, Randn}。

Background

原框架随机初始化依赖局部 std::mt19937，缺少统一后端抽象、默认随机源、全局 seed 和可恢复 state，无法稳定支持显式独立随机流、跨运行复现和 CUDA kernel 间不重叠的随机区间。

Main Changes

Generator 基础设施

• 新增用户侧 Generator 句柄和多态 GeneratorImpl。
• 实现 CPUGeneratorImpl 和 CUDAGeneratorImpl。
• 支持 ManualSeed、ManualSeedAll、Seed、InitialSeed、GetState、SetState 和设备查询。
• CPU 保存完整 mt19937 engine state；CUDA state v3 保存 seed、来源设备和 semantic offset。
• Generator 内部状态访问和 CUDA offset reservation 具备线程安全保护。

默认实例与算子接入

• CPU 使用进程级默认 Generator；CUDA 按 device index 维护独立默认实例。
• 未显式传入 Generator 时，按 tensor/device 自动选择默认实例。
• 显式 Generator 路径不会推进默认 Generator 状态。
• CUDA Uniform/Normal 使用项目内 Philox-style seed/offset kernel，并在调用前预留不重叠随机区间。

测试

• 新增 CPU Generator 41 项测试和 CUDA Generator 16 项测试。
• 覆盖同/异 seed、state 重放、默认/显式路径、损坏 state、线程安全、随机流连续性、空 tensor、参数边界和双物理 GPU 隔离。
• 新增跨进程复现脚本，用于比较固定 seed 下的稳定 digest。

Compatibility and Scope

• 不传随机源的既有调用方式保持不变。
• 显式随机源参数由 std::optional<std::mt19937> 迁移为 std::shared_ptr<Generator>；外部显式调用方需迁移。
• CPU/CUDA 公共 Uniform/Normal 当前明确限定为 float32。
• 不要求与 PyTorch 逐 bit 一致，也不要求 CPU 与 CUDA 输出一致。
• 当前框架没有成熟 Dropout 算子；本 PR 以 Rand/Randn 完成训练期随机调用接入，不提交缺少 training/eval、mask scaling 和 autograd 语义的半成品 Dropout。
• 示例 checkpoint loader 的局部 mt19937 不在本次核心随机调用链范围内。

Tests

验证环境：2× NVIDIA GeForce RTX 2080 Ti、CUDA 12.4、GCC/G++ 13.4.0、CMake 3.28.4。

Test	Result
CPU Generator	41/41 passed
CPU label regression	231 passed, 11 skipped, 0 failed
CUDA Generator	16/16 passed
CUDA label regression	9/9 passed
Dual physical GPU isolation	10/10 passed
Cross-process reproducibility	digest ec0f626ebd167804 matched
ASan + UBSan	41/41 passed, no sanitizer error
Format / whitespace	passed

Test Screenshots

• 01_cpu_generator_tests.png

• 02_cuda_generator_tests.png

Reproduction

cmake -S . -B build-cpu -G Ninja -DUSE_CUDA=OFF -DUSE_NCCL=OFF -DBUILD_TEST=ON
cmake --build build-cpu
build-cpu/tests/tensor/test_tensor_cpu_only --gtest_filter='CPUGeneratorTest.*'
scripts/check_generator_reproducibility.sh build-cpu/tests/tensor/test_tensor_cpu_only

cmake -S . -B build-cuda -G Ninja -DUSE_CUDA=ON -DUSE_NCCL=OFF -DBUILD_TEST=ON
cmake --build build-cuda
build-cuda/tests/tensor/test_tensor_cuda_only --gtest_filter='CUDAGeneratorTensorTest.*'

python3 scripts/format.py --path infini_train example --check

AI Assistance Disclosure

本项目由参赛者主导并实际参与，使用 OpenAI Codex 作为辅助开发工具。具体说明见随 PR 上传的 REFERENCE.md；全部验收日志均来自目标环境真实执行。

[Copilot]

Pull request overview

This PR introduces a unified random number generation infrastructure (Generator / GeneratorImpl) across CPU and CUDA backends, enabling reproducible seeding, per-device default generators, state save/restore, and consistent integration into tensor random APIs and initialization utilities.

Changes:

• Added Generator handle + polymorphic backend implementations (CPU mt19937 state, CUDA Philox-style seed/offset with thread-safe offset reservation) and default generator management.
• Wired the generator flow into Tensor::Uniform, nn::init::{Uniform, Normal, KaimingUniform}, and nn::function::{Rand, Randn}, including CUDA kernels for float32 uniform/normal.
• Added extensive CPU/CUDA generator tests and a script to validate cross-process reproducibility digest.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/tensor/cuda_only/test_generator_cuda.cc	New CUDA-only tests validating default/explicit generator behavior, state replay, offset semantics, and kernel statistical sanity.
tests/tensor/cpu_only/test_generator.cc	New CPU-only tests covering generator API/state serialization, parameter validation, default-generator semantics, and CUDA-generator host-side offset logic.
scripts/check_generator_reproducibility.sh	New helper script to run a reproducibility digest test twice and compare outputs across processes.
infini_train/src/tensor.cc	Updates Tensor::Uniform to accept std::shared_ptr<Generator> and route through nn::init::Uniform.
infini_train/src/nn/parallel/ddp/param_and_grad_buffer.cc	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/nn/init.cc	Refactors Uniform/Normal/KaimingUniform to use Generator, adds bounds/parameter validation, and dispatches CUDA float32 random kernels with reserved offsets.
infini_train/src/nn/functional.cc	Adds nn::function::Rand and Randn that create float32 tensors and delegate to init random ops with optional generator.
infini_train/src/kernels/cuda/reduction.cu	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/kernels/cuda/random.cu	New CUDA Philox-style random uniform/normal float32 kernels and launcher functions.
infini_train/src/kernels/cuda/no_op.cu	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/kernels/cuda/gather.cu	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/kernels/cuda/elementwise.cu	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/kernels/cpu/transform.cc	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/kernels/cpu/cross_entropy.cc	Adds missing <numeric> include for existing std::accumulate usage.
infini_train/src/generator.cc	New core implementation of CPU/CUDA generators, default-generator management, seeding APIs, and state serialization/parsing.
infini_train/include/tensor.h	Updates Tensor::Uniform signature to take std::shared_ptr<Generator> (default nullptr) and forward-declares Generator.
infini_train/include/nn/init.h	Updates init APIs to accept std::shared_ptr<Generator> and includes generator definitions.
infini_train/include/nn/functional.h	Declares Rand/Randn APIs with device + optional generator parameters and necessary forward declarations/includes.
infini_train/include/generator.h	New public generator interface (impl vtable, handle, default-generator APIs, seeding/state APIs).
infini_train/include/common/cuda/kernel_helper.cuh	Adjusts Tanh implementation for nv_bfloat16 / half via float conversion.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[surprisely]

HONOR_CODE.md
REFERENCE.md
赛题报告.pdf