Results for “mixed-precision”

Evaluates geometric consistency in text-to-video generation by measuring structural and motion coherence across camera trajectories, detecting deformation and occlusion artifacts in static scenes.

Provides practical guidance for training Mixture-of-Experts Vision-Language Models in Megatron Bridge, comparing FSDP and 3D-parallel approaches with lessons from recent multimodal experiments.

Provides guidance for training Mixture-of-Experts models with long context windows, covering context parallelism sizing, selective recomputation, dispatcher choices, and practical patterns from recent experiments.

Trace and interpret the Pareto frontier across competing objectives using repeated single-objective cuOpt solves (weighted-sum and ε-constraint).

Implements Manifold-Constrained Hyper-Connections (mHC) using Doubly Stochastic Matrices to improve deep learning stability.

Evaluates automated metrics and vision-language models on identifying granular errors in detailed image descriptions and ranking paired descriptions against human judgments, using macro F1, pairwise accuracy, Spearman rank ρ, and Kendall's τ.

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace, covering architectures, routing, load balancing, and expert parallelism.

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques for 1.5-3.6× speedup without quality loss.

Quantifies the minimum accuracy gap needed between two models for a sampled micro-benchmark to reliably preserve their ranking, using the MDAD metric from Yauney et al. (2025).

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.

Evaluates computational throughput and real-time efficiency of embedded CPU and GPU platforms by measuring peak FLOPS via a matrix rotation kernel and assessing inference latency and power consumption on a robotic vision pipeline.

Provides guidance on using PyTorch for deep learning, covering tensors, autograd, nn.Module, DataLoaders, and best practices.

Provides expert guidance for distributed training with DeepSpeed, covering ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, and sparse attention.