ascend-robotdocs(pytorch): updated incorrect link

MindSpeed LLM Documentation Guide

Documentation Overview

The MindSpeed LLM documentation is organized by training framework and mainly includes the following core directories:

pytorch/: Documentation based on the PyTorch training framework. It mainly supports the MCore and FSDP2 training backends and includes installation guides, model lists, feature descriptions, training solutions, and toolchains.
mindspore/: Documentation based on the MindSpore training framework. It supports only the MCore training backend and provides usage guides and feature descriptions for the MindSpore framework.

Documentation Directory Structure

The MindSpeed LLM documentation directory hierarchy is shown below:

docs/zh/

├── introduction.md         # Project introduction
├── project_guide.md        # Project guide
├── docs_guide.md           # Documentation guide
├── appendixes.md           # Appendix documents
├── pytorch/                # Documentation related to the PyTorch training framework
│   ├── develop/            # Development guide
│   │   ├── mcore/          # MCore development guide
│   │   │   └── lora_finetune_adaptation.md # LoRA fine-tuning migration development
│   │   └── fsdp2/          # FSDP2 development guide
│   │       └── model_adaptation.md # FSDP2 model adaptation
│   ├── features/           # Feature documents
│   │   ├── mcore/          # MCore feature documents
│   │   └── fsdp2/          # FSDP2 feature documents
│   │       ├── arguments.md            # FSDP2 parameter descriptions
│   │       └── fsdp2_basic_features.md # FSDP2 feature descriptions
│   ├── figures/            # Images
│   ├── models/             # Models supported by the PyTorch framework
│   │   └── supported_models.md
│   ├── training/           # Training solution documents
│   │   ├── install_guide.md  # Installation guide
│   │   ├── quick_start.md    # Quick start guide
│   │   ├── evaluation/       # Model evaluation
│   │   │   ├── evaluation_guide.md
│   │   │   ├── models_evaluation.md
│   │   │   └── evaluation_datasets/  # Evaluation datasets
│   │   ├── finetune/       # Model fine-tuning
│   │   │   ├── mcore/      # MCore fine-tuning solutions
│   │   │   └── fsdp2/      # FSDP2 fine-tuning solutions
│   │   │       └── finetune_fsdp2.md
│   │   ├── inference/      # Model inference
│   │   │   ├── inference.md
│   │   │   └── chat.md
│   │   └── pretrain/       # Model pretraining
│   │       └── mcore/      # MCore pretraining solutions
│   │           ├── pretrain.md
│   │           ├── pretrain_eod.md
│   │           └── train_from_hf.md
│   └── tools/              # Tool documents
│       ├── data_process_sft_alpaca_style.md   # Alpaca-style data processing
│       ├── data_process_sft_sharegpt_style.md # ShareGPT-style data processing
│       ├── data_process_dpo_pairwise.md       # Pairwise data processing
│       ├── data_process_pretrain.md           # Pretraining data processing
│       ├── checkpoint_convert_hf_mcore_large_params.md  # Weight conversion V2
│       ├── checkpoint_convert_hf_dcp.md       # Hugging Face-DCP weight conversion
│       ├── profiling.md                       # Performance analysis
│       └── deterministic_computation.md       # Deterministic computation
└── mindspore/              # Documentation related to the MindSpore training framework
    ├── readme.md           # MindSpore documentation overview
    ├── quick_start.md      # Quick start guide
    ├── install_guide.md    # Installation guide
    ├── features/           # MindSpore feature documents
    └── models/             # Models supported by the MindSpore framework

Quick links: Getting Started | MCore Backend | FSDP2 Backend | Toolchain

Getting Started

Content	Description
install_guide_pytorch	Installation guidance for the PyTorch framework environment.
quick_start_pytorch	Quick start guidance for the MCore backend, covering the full process from environment setup to model pretraining and fine-tuning on the PyTorch framework.
install_guide_mindspore	Installation guidance for the MindSpore framework environment.
quick_start_mindspore	Quick start guidance for the MCore backend, covering the full process from environment setup to model pretraining and fine-tuning on the MindSpore framework.
finetune_fsdp2	Quick start guidance for the FSDP2 backend, covering the full process from environment setup to model training.
supported_models	Model support list.

MCore Backend

Features

Content	Description
features	A collection of performance optimization and memory optimization features supported by parts of the repository.

Development Guide

Content	Description
lora_finetune_adaptation	LoRA fine-tuning migration development guide.

Training Solutions

Category	Content	Description
Pretraining	pretrain	Multi-sample pretraining method.
	pretrain_eod	Multi-sample pack pretraining method.
Fine-tuning	instruction_finetune	Full-parameter model fine-tuning solution.
	multi_sample_pack_finetune	Multi-sample pack fine-tuning solution.
	multi_turn_conversation	Multi-turn conversation fine-tuning solution.
	lora_finetune	LoRA model fine-tuning solution.
	qlora_finetune	QLoRA model fine-tuning solution.
Inference	inference	Model inference.
	chat	Chat.
	yarn	Uses the Yarn solution to extend context length and support long-sequence inference.
Evaluation	evaluation_guide	Model evaluation solution.
	models_evaluation	Repository model evaluation list.
	evaluation_datasets	Evaluation datasets supported by the repository.

FSDP2 Backend