training-llms-megatron
Framework for large-scale distributed LLM training with high hardware utilization on NVIDIA GPUs.
Install
mkdir -p .claude/skills/training-llms-megatron && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/7702" && unzip -o skill.zip -d .claude/skills/training-llms-megatron && rm skill.zipInstalls to .claude/skills/training-llms-megatron
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.Key capabilities
- →Orchestrates distributed 3D parallelism (tensor, pipeline, sequence)
- →Optimizes GPU memory usage for models >1B parameters
- →Launches multi-node training jobs
- →Monitors H100 efficiency metrics
How it works
Manages communication primitives between GPU groups to shard parameters and gradients across a distributed cluster.
Inputs & outputs
When to use training-llms-megatron
- →Configure model parallelism
- →Launch distributed training
- →Monitor training performance
About this skill
Megatron-Core - Large-Scale LLM Training
Quick start
Megatron-Core trains LLMs from 2B to 462B parameters with up to 47% Model FLOP Utilization on H100 GPUs through advanced parallelism strategies.
Installation:
# Docker (recommended)
docker run --gpus all -it --rm nvcr.io/nvidia/pytorch:25.04-py3
# Or pip
pip install megatron-core
Simple distributed training:
# Train with 2 GPUs using data parallelism
torchrun --nproc_per_node=2 examples/run_simple_mcore_train_loop.py
# Or LLaMA-3 8B training
./examples/llama/train_llama3_8b_fp8.sh
Common workflows
Workflow 1: Train LLaMA-style model with 3D parallelism
Copy this checklist:
LLaMA Training Setup:
- [ ] Step 1: Choose parallelism configuration
- [ ] Step 2: Configure training hyperparameters
- [ ] Step 3: Launch distributed training
- [ ] Step 4: Monitor performance metrics
Step 1: Choose parallelism configuration
Model size determines parallelism strategy:
| Model Size | GPUs | Tensor Parallel | Pipeline Parallel | Data Parallel | Context Parallel |
|---|---|---|---|---|---|
| 7B | 8 | 1 | 1 | 8 | 1 |
| 13B | 8 | 2 | 1 | 4 | 1 |
| 70B | 64 | 4 | 4 | 4 | 1 |
| 405B | 128 | 8 | 8 | 2 | 2 |
Step 2: Configure training hyperparameters
#!/bin/bash
# train_llama_70b.sh
GPUS_PER_NODE=8
NNODES=8 # 64 GPUs total
TP=4 # Tensor parallel
PP=4 # Pipeline parallel
CP=1 # Context parallel
# LLaMA 70B configuration
MODEL_SIZE=70 # Billion parameters
HIDDEN_SIZE=8192
NUM_LAYERS=80
NUM_HEADS=64
SEQ_LENGTH=4096
# Training hyperparameters
MICRO_BATCH=1
GLOBAL_BATCH=1024
LR=3e-4
torchrun \
--nproc_per_node=$GPUS_PER_NODE \
--nnodes=$NNODES \
pretrain_gpt.py \
--tensor-model-parallel-size $TP \
--pipeline-model-parallel-size $PP \
--context-parallel-size $CP \
--sequence-parallel \
--num-layers $NUM_LAYERS \
--hidden-size $HIDDEN_SIZE \
--num-attention-heads $NUM_HEADS \
--seq-length $SEQ_LENGTH \
--max-position-embeddings $SEQ_LENGTH \
--micro-batch-size $MICRO_BATCH \
--global-batch-size $GLOBAL_BATCH \
--lr $LR \
--train-iters 100000 \
--lr-decay-style cosine \
--lr-warmup-iters 2000 \
--weight-decay 0.1 \
--clip-grad 1.0 \
--bf16 \
--use-mcore-models \
--transformer-impl transformer_engine \
--data-path /path/to/data \
--vocab-file /path/to/vocab.json \
--merge-file /path/to/merges.txt
Step 3: Launch distributed training
# Single node (8 GPUs)
bash train_llama_70b.sh
# Multi-node with SLURM
sbatch --nodes=8 --gpus-per-node=8 train_llama_70b.sh
Step 4: Monitor performance metrics
Key metrics to track:
Model FLOP Utilization (MFU): Target >40% on H100
Throughput: Tokens/sec/GPU
Memory usage: <80GB per GPU for 70B model
Loss: Should decrease steadily
Workflow 2: Configure Mixture of Experts (MoE) training
For sparse MoE models like Mixtral.
MoE Training:
- [ ] Step 1: Configure expert parallelism
- [ ] Step 2: Set MoE hyperparameters
- [ ] Step 3: Launch training with EP
Step 1: Configure expert parallelism
# Mixtral 8x7B example
TENSOR_PARALLEL=2
PIPELINE_PARALLEL=1
EXPERT_PARALLEL=4 # Split 8 experts across 4 GPUs
DATA_PARALLEL=4
TOTAL_GPUS=$((TENSOR_PARALLEL * PIPELINE_PARALLEL * EXPERT_PARALLEL * DATA_PARALLEL))
# = 2 * 1 * 4 * 4 = 32 GPUs
Step 2: Set MoE hyperparameters
torchrun \
--nproc_per_node=8 \
pretrain_gpt.py \
--tensor-model-parallel-size 2 \
--pipeline-model-parallel-size 1 \
--expert-model-parallel-size 4 \
--num-experts 8 \
--moe-router-topk 2 \
--moe-router-load-balancing-type aux_loss \
--moe-aux-loss-coeff 0.01 \
--hidden-size 4096 \
--num-layers 32 \
--num-attention-heads 32 \
--seq-length 4096 \
--max-position-embeddings 4096 \
--bf16 \
--use-mcore-models \
--transformer-impl transformer_engine \
--data-path /path/to/data \
--vocab-file /path/to/vocab.json \
--merge-file /path/to/merges.txt
Step 3: Launch training with EP
Expert parallelism distributes different experts across GPUs, reducing memory while maintaining capacity.
Memory without EP: 8 experts × 7B = 56GB per GPU
Memory with EP=4: 2 experts × 7B = 14GB per GPU
Savings: 75% memory reduction
Workflow 3: Optimize for maximum throughput
Achieve 47% MFU on H100.
Performance Optimization:
- [ ] Step 1: Enable Flash Attention
- [ ] Step 2: Use FP8 precision (H100)
- [ ] Step 3: Optimize micro-batch size
- [ ] Step 4: Tune parallelism degrees
Step 1: Enable optimizations
--use-mcore-models # Use Megatron Core models
--transformer-impl transformer_engine # Use Transformer Engine
--sequence-parallel # Reduce activation memory (use with TP)
Step 2: Use FP8 precision (H100 only)
--fp8-hybrid # FP8 mixed precision training
# Transformer Engine handles FP8 automatically
Result: 1.5-2x speedup on H100 vs BF16.
Step 3: Optimize micro-batch size
Find largest micro-batch that fits in memory:
# Start with 1, increase until OOM
for MBS in 1 2 4 8; do
echo "Testing micro-batch-size=$MBS"
torchrun ... --micro-batch-size $MBS
done
Typical values:
- 7B model: 4-8
- 70B model: 1-2
- 405B model: 1
Step 4: Tune parallelism degrees
Rules of thumb:
Tensor Parallel: Use ≤8 (limited by NVLink within node)
Pipeline Parallel: Use for >70B models
Context Parallel: Use for sequences >8K tokens
Data Parallel: Fill remaining GPUs
Example 405B on 128 H100s:
TP=8 (1 node)
PP=8 (across nodes)
CP=2 (long sequences)
DP=1
Total = 8 × 8 × 2 × 1 = 128 GPUs
When to use vs alternatives
Use Megatron-Core when:
- Training models >10B parameters
- Need maximum efficiency (target >40% MFU)
- Using NVIDIA GPUs (A100, H100)
- Production training at scale
- Want fine-grained parallelism control
Use alternatives instead:
- PyTorch FSDP: Models <70B, simpler API, PyTorch native
- DeepSpeed: Easier setup, good for <100B models
- HuggingFace Accelerate: Prototyping, simpler workflows
- LitGPT: Educational, single-file implementations
Common issues
Issue: Low GPU utilization (<30% MFU)
Causes:
- Micro-batch too small
- Too much parallelism overhead
- Not using Flash Attention
Fixes:
# Increase micro-batch
--micro-batch-size 4 # Was 1
# Enable optimizations
--use-flash-attn
--sequence-parallel
# Reduce TP if >8
--tensor-model-parallel-size 4 # Was 16
Issue: Out of memory
Reduce memory with:
--tensor-model-parallel-size 2 # Split model across GPUs
--recompute-granularity full # Gradient checkpointing
--recompute-method block # Checkpoint transformer blocks
--recompute-num-layers 1 # Checkpoint every layer
Or use CPU/NVMe offloading:
--cpu-optimizer # Offload optimizer to CPU
--cpu-optimizer-type ADAM # CPU Adam variant
Issue: Training slower than expected
Check:
- Network bottleneck: Ensure InfiniBand/NVLink enabled
- Pipeline bubbles: Use interleaved pipeline schedule
--num-layers-per-virtual-pipeline-stage 2 - Data loading: Use fast data loader
--dataloader-type cyclic
Issue: Diverging loss
Stabilize training:
--lr-warmup-iters 2000 # Longer warmup
--clip-grad 1.0 # Gradient clipping
--init-method-std 0.006 # Smaller init
--attention-dropout 0.0 # No dropout in attention
--hidden-dropout 0.0 # No dropout in FFN
Advanced topics
Parallelism strategies: See references/parallelism-guide.md for detailed comparison of TP/PP/DP/CP/EP with performance analysis and when to use each.
Performance benchmarks: See references/benchmarks.md for MFU numbers across different model sizes and GPU configurations.
Production configurations: See references/production-examples.md for real-world setups from LLaMA 3 405B, Nemotron-4 340B, and DeepSeek-V3 671B.
Training recipes: See references/training-recipes.md for complete hyperparameter configurations for GPT/LLaMA/Mixtral architectures.
Hardware requirements
- GPU: NVIDIA Ampere+ (A100, H100, B200)
- Turing works but slower
- FP8 requires Hopper/Ada/Blackwell
- Network: InfiniBand or 400Gb+ Ethernet for multi-node
- Memory per GPU:
- 7B model: 40GB+
- 70B model: 80GB (with TP=4)
- 405B model: 80GB (with TP=8, PP=8)
- Storage: Fast NVMe for checkpoints (1TB+ for 70B+ models)
Resources
- Docs: https://docs.nvidia.com/megatron-core/
- GitHub: https://github.com/NVIDIA/Megatron-LM
- Papers:
- "Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism" (2019)
- "Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM" (2021)
- NeMo Framework: https://docs.nvidia.com/nemo-framework/ (built on Megatron-Core)
When not to use it
- →Small models that fit on single GPUs
- →Prototyping without the need for distributed infrastructure
Prerequisites
Limitations
- →High infrastructure requirement for full parallelism usage
- →Steep configuration learning curve for distributed settings
How it compares
Handles the complex boilerplate of sharding models across thousands of GPUs, achieving higher FLOP efficiency than standard frameworks.
Compared to similar skills
training-llms-megatron side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| training-llms-megatron (this skill) | 1 | 7mo | Review | Advanced |
| mlflow | 3 | 7mo | Review | Intermediate |
| modal-serverless-gpu | 1 | 7mo | Review | Intermediate |
| hugging-face-model-trainer | 1 | 6mo | Review | Advanced |
Try saying
Example prompts that trigger this skill in your AI assistant.
More by davila7
View all by davila7 →You might also like
mlflow
davila7
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform
modal-serverless-gpu
davila7
Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as APIs, or running batch jobs with automatic scaling.
hugging-face-model-trainer
patchy631
This skill should be used when users want to train or fine-tune language models using TRL (Transformer Reinforcement Learning) on Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV scripts with PEP 723 format, dataset preparation and validation, hardware selection, cost estimation, Trackio monitoring, Hub authentication, and model persistence. Should be invoked for tasks involving cloud GPU training, GGUF conversion, or when users mention training on Hugging Face Jobs without local GPU setup.
tensorrt-llm
davila7
Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4), in-flight batching, and multi-GPU scaling.
flash
MachinaIO
runpod-flash SDK and CLI for deploying AI workloads on Runpod serverless GPUs/CPUs.
building-automl-pipelines
jeremylongshore
Build automated machine learning pipelines, including feature engineering, model selection, and performance evaluation.