Results for “nemo”

Guides implementation of new model architectures in NeMo AutoModel through five phases: discovery, implementation, registration, validation, and testing.

Generates custom safety policies for NVIDIA Nemotron content-safety guardrails, producing a Markdown policy, JSON taxonomy, and inference prompts from rough user input.

Launch, monitor, stop, and debug NeMo-RL recipes on a Kubernetes cluster using the nrl-k8s CLI, supporting ephemeral and long-lived RayCluster modes.

Provides storage and environment conventions for NeMo-RL agents on Brev instances, ensuring large experiment outputs go to /ephemeral and secrets are loaded from .env.

Plan, configure, and chain Nemotron model customization steps into single-step or multi-step pipelines for curation, translation, fine-tuning, RL alignment, benchmarking, checkpoint conversion, optimization, and evaluation.

Guides agents through the full lifecycle of NeMo-RL experiments: understanding recipes, launching reproducible runs, analyzing results, and preserving human oversight with git and TSV logs.

Build synthetic datasets and data generation pipelines using the Data Designer library.

Indexes Megatron Bridge recipes and recommends the best starting config based on model, GPU count, and training goal.

Routes NVIDIA Nemotron Speech (Riva) NIM tasks for ASR, TTS, and NMT, covering cloud-hosted inference, self-hosted Docker deployment, and custom model builds.

Configure fault tolerance, straggler detection, preemption, in-process restart, and re-run state machine for Megatron Bridge training jobs.

Run Megatron-LM (MLM) and Megatron Bridge training with mock or real data, covering correlation testing, available recipes, and multi-GPU examples.

Curates clinical-specialty term lists, generates IPA-tagged synthetic audio via TTS, and produces NeMo-format manifests for ASR benchmark evaluation.

Index folders of PDFs and other documents into LanceDB for vector search, then query them with semantic search, page filters, verbatim quotes, and cross-document aggregation.

Configure and validate CPU offloading for Megatron Bridge training, including activation offloading and optimizer state offloading with HybridDeviceOptimizer.

Maintain durable session memory across agent disconnects by writing structured checkpoints to the repo's session directory, enabling context recovery.

Navigate the PhysicsNeMo repository by discovering model families, datapipes, and examples through live file search, without writing training code.

Convert single-node PyTorch distributed scripts into multi-node Slurm sbatch jobs and debug common multi-node failures, covering srun-native and torch.distributed approaches, container setup, NCCL timeouts, and interactive allocation.

Validate and configure packed sequences and long-context training in Megatron-Bridge, distinguishing offline packed SFT for LLMs from in-batch packing for VLMs with correct context parallelism constraints.

Enables Megatron Fully Sharded Data Parallel in Megatron-Bridge with configuration overrides, code anchors, pitfalls, and verification steps.

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.

Optimizes MoE expert-parallel communication overlap in Megatron Bridge, covering dispatch/combine overlap, flex dispatcher backends, and expert wgrad scheduling.

Validate and use CUDA graph capture in Megatron Bridge, including local full-iteration graphs and Transformer Engine scoped graphs for attention, MLP, and MoE modules.

Reduces peak GPU memory in Megatron Bridge training by applying expandable segments, parallelism resizing, activation recompute, and CPU offloading constraints.

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.