Skip to content

CLAUDE.md

Latest commit

 

History

History
86 lines (69 loc) · 3.28 KB

CLAUDE.md

File metadata and controls

86 lines (69 loc) · 3.28 KB

CLAUDE.md - LMMs Engine Repository Guide

Repository Overview

LMMs Engine is a training framework for Large Multimodal Models (LMMs) developed by LMMs-Lab. This is a Python package focused on highly efficient training of multimodal models with support for various architectures and training paradigms.

Configuration System

Training is configured via YAML files. YAML is recommended for better readability and comment support.

YAML Configuration (Recommended)

# Configuration supports comments in YAML
- type: trainer
  config:
    trainer_type: hf_trainer
    dataset_config:
      # Dataset configuration
      dataset_name: "example_dataset"
    model_config:
      # Model configuration  
      model_name: "example_model"
    # TrainingArguments parameters
    output_dir: "./output"
    num_train_epochs: 3

Development Commands

Training Launch

# Recommended: torchrun with YAML config
torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
  --master_addr="127.0.0.1" --master_port="8000" \
  -m lmms_engine.launch.cli config_yaml=path/to/config.yaml

# Hydra overrides (no config file needed)
torchrun --nproc_per_node=8 --nnodes=1 --node_rank=0 \
  --master_addr="127.0.0.1" --master_port="8000" \
  -m lmms_engine.launch.cli \
  trainer_type=fsdp2_trainer \
  model_config.load_from_pretrained_path="model/path" \
  trainer_args.output_dir="./output" \
  trainer_args.num_train_epochs=3

Development Setup

# Install with development dependencies
uv pip install -e ".[all]"  # Includes preference learning and storage

# Performance optimizations
uv pip install flash-attn --no-build-isolation
uv pip install liger-kernel

Development Philosophy

  • Simplicity: Write simple, straightforward code
  • Readability: Make code easy to understand
  • Performance: Consider performance without sacrificing readability
  • Maintainability: Write code that's easy to update
  • Testability: Ensure code is testable
  • Reusability: Create reusable components and functions
  • Less Code = Less Debt: Minimize code footprint

Coding Best Practices

  • Early Returns: Use to avoid nested conditions
  • Descriptive Names: Use clear variable/function names (prefix handlers with "handle")
  • Constants Over Functions: Use constants where possible
  • DRY Code: Don't repeat yourself
  • Functional Style: Prefer functional, immutable approaches when not verbose
  • Minimal Changes: Only modify code related to the task at hand
  • Function Ordering: Define composing functions before their components
  • TODO Comments: Mark issues in existing code with "TODO:" prefix
  • Simplicity: Prioritize simplicity and readability over clever solutions
  • Build Iteratively Start with minimal functionality and verify it works before adding complexity
  • Run Tests: Test your code frequently with realistic inputs and validate outputs
  • Build Test Environments: Create testing environments for components that are difficult to validate directly
  • Functional Code: Use functional and stateless approaches where they improve clarity
  • Clean logic: Keep core logic clean and push implementation details to the edges
  • File Organsiation: Balance file organization with simplicity - use an appropriate number of files for the project scale