Mixture of Experts: Core Concepts and Hands-on Implementation
Chapter 1: Foundations of Mixture of Experts Models
Overview of Sparsely-Gated MoE Architecture
The Gating Network: Formulation and Function
Expert Networks: Specialization and Capacity
Mathematical Formulation of the MoE Layer
Load Balancing and Auxiliary Losses
Challenges in MoE Training: Expert Collapse
Comparison with Dense Model Scaling
Hands-on: Implementing a Basic MoE Layer
Chapter 2: Advanced Routing Mechanisms
Analysis of Top-k Gating and its Variants
Noisy Top-k Gating for Load Balancing
Hash-based Routing for Deterministic Selection
Switch Transformers: Simplified Routing
Soft MoE: Differentiable Routing
Analyzing Routing Decisions and Specialization
Hands-on: Implementing Different Routing Strategies
Chapter 3: Training and Optimization of Large-Scale MoEs
Expert Parallelism for Distributed Training
Combining Model, Data, and Expert Parallelism
Capacity Factor and its Impact on Performance
Techniques for Mitigating Router Z-Loss Instability
Precision and its Effects: BFloat16 Training
Fine-tuning Strategies for Pre-trained MoE Models
Practice: Configuring a Distributed Training Job
Chapter 4: Efficient Inference with MoE Models
Inference Challenges: Memory and Latency
Expert Offloading to CPU or NVMe
Batching Strategies for Sparse Activation
Model Distillation for MoE Compression
Quantization Techniques for MoE Layers
Speculative Decoding with MoE Models
Hands-on: Building an Optimized Inference Pipeline
Chapter 5: Integrating MoE into Modern Architectures
Replacing FFNs with MoE Layers in Transformers
Placement of MoE Layers: Frequency and Location
MoE in Vision Transformers (ViT)
MoE in Multi-modal Models
Architectural Variants and their Properties
Analyzing Parameter vs. FLOPs Trade-offs
Practice: Modifying a Transformer to use MoE
Quantization Techniques for MoE Layers
Was this section helpful?
- GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers, Elias Frantar, Saleh Ashkboos, Torsten Hoefler, Dan Alistarh, 2022 arXiv preprint arXiv:2210.17323 DOI: 10.48550/arXiv.2210.17323 - Introduces GPTQ, a post-training quantization method for large language models, relevant for quantizing experts.
- QLoRA: Efficient Finetuning of Quantized LLMs on Consumer GPUs, Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, Luke Zettlemoyer, 2023 arXiv preprint arXiv:2305.14314 DOI: 10.48550/arXiv.2305.14314 - Presents NF4 data type and 4-bit quantization, enabling efficient finetuning of large language models.
- LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale, Tim Dettmers, Mike Lewis, Younes Belkada, Luke Zettlemoyer, 2022 NeurIPS 2022 DOI: 10.48550/arXiv.2208.07339 - Describes 8-bit matrix multiplication and quantization techniques used in the
bitsandbyteslibrary for large language models. - AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration, Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, Song Han, 2023 MLSys 2024 DOI: 10.48550/arXiv.2306.00978 - Proposes AWQ, an activation-aware weight quantization method that maintains model performance by protecting important weights.
- Quantization (bitsandbytes), Hugging Face, 2024 (Hugging Face documentation) - Official documentation explaining how to apply quantization with
bitsandbyteswithin the Hugging Face Transformers library.