
TorchTitan
FreeEfficient distributed LLM pretraining with PyTorch.
Free · Opens the source repo
What TorchTitan does
TorchTitan is a robust framework designed for efficient large-scale pretraining of language models using PyTorch. It leverages advanced 4D parallelism techniques, including Fully Sharded Data Parallel (FSDP2), Tensor Parallelism (TP), Pipeline Parallelism (PP), and Context Parallelism (CP). This combination allows users to achieve significant speed improvements—over 65% on H100 GPUs—making it suitable for training models ranging from Llama 3.1 to custom architectures across 8 to 512+ GPUs. The framework is particularly beneficial for those looking to implement Float8 precision, which can enhance performance on compatible hardware.
Users can easily get started with TorchTitan by installing it via PyPI or from source for the latest features. The setup process includes downloading necessary tokenizer assets from Hugging Face and configuring training parameters through TOML files. The skill supports various workflows, including single-node and multi-node training, enabling users to scale their training efforts based on their infrastructure. This flexibility is crucial for researchers and engineers aiming to push the boundaries of language model capabilities.
TorchTitan also provides comprehensive documentation and common workflows that guide users through the pretraining process, including monitoring and checkpointing. For advanced users, it offers detailed configuration options for optimizing parallelism and memory usage, ensuring that even large models can be trained efficiently without running into memory issues. The framework's focus on interoperability with existing PyTorch tools and libraries makes it a valuable addition to any developer's toolkit.
When to use it
Use TorchTitan when you need a PyTorch-native solution for pretraining language models at scale, especially when leveraging advanced parallelism techniques.
When not to use it
This skill may not be suitable for users looking for fine-tuning capabilities or those who require extensive support for NVIDIA-specific optimizations found in other frameworks.
What you can build with it
Pretraining Llama 3.1 8B
Set up a single-node training environment to pretrain the Llama 3.1 model using 8 GPUs.
Multi-node Training with SLURM
Configure and submit a SLURM job script to train large models across multiple nodes.
Enabling Float8 Training
Optimize training on H100 GPUs by enabling Float8 precision for faster computation.
How to install TorchTitan
View source1. Install with the skills CLI
npx skills add davila7/claude-code-templates/model-architecture-torchtitan --agent claude-code2. Or install it manually
Download the skill folder and drop it into ~/.claude/skills/ for all projects, or .claude/skills/ to scope it to one repo. Restart Claude Code so it picks up the new skill.
Anthropic's agentic coding CLI, and the reference implementation of Agent Skills. Drop a skill folder into ~/.claude/skills and Claude Code loads it automatically whenever a task matches the skill's description. Claude Code docs
Inside SKILL.md
Written by davila7TorchTitan - PyTorch Native Distributed LLM Pretraining
Quick start
TorchTitan is PyTorch's official platform for large-scale LLM pretraining with composable 4D parallelism (FSDP2, TP, PP, CP), achieving 65%+ speedups over baselines on H100 GPUs.
Installation:
# From PyPI (stable)
pip install torchtitan
# From source (latest features, requires PyTorch nightly)
git clone https://github.com/pytorch/torchtitan
cd torchtitan
pip install -r requirements.txt
Download tokenizer:
# Get HF token from https://huggingface.co/settings/tokens
python scripts/download_hf_assets.py --repo_id meta-llama/Llama-3.1-8B --assets tokenizer --hf_token=...
Start training on 8 GPUs:
CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh
Common workflows
Workflow 1: Pretrain Llama 3.1 8B on single node
Copy this checklist:
Single Node Pretraining:
- [ ] Step 1: Download tokenizer
- [ ] Step 2: Configure training
- [ ] Step 3: Launch training
- [ ] Step 4: Monitor and checkpoint
Step 1: Download tokenizer
python scripts/download_hf_assets.py \
--repo_id meta-llama/Llama-3.1-8B \
--assets tokenizer \
--hf_token=YOUR_HF_TOKEN
Step 2: Configure training
Edit or create a TOML config file:
# llama3_8b_custom.toml
[job]
dump_folder = "./outputs"
description = "Llama 3.1 8B training"
[model]
name = "llama3"
flavor = "8B"
hf_assets_path = "./assets/hf/Llama-3.1-8B"
[optimizer]
name = "AdamW"
lr = 3e-4
[lr_scheduler]
warmup_steps = 200
[training]
local_batch_size = 2
seq_len = 8192
max_norm = 1.0
steps = 1000
dataset = "c4"
[parallelism]
data_parallel_shard_degree = -1 # Use all GPUs for FSDP
[activation_checkpoint]
mode = "selective"
selective_ac_option = "op"
[checkpoint]
enable = true
folder = "checkpoint"
interval = 500
Step 3: Launch training
# 8 GPUs on single node
CONFIG_FILE="./llama3_8b_custom.toml" ./run_train.sh
# Or explicitly with torchrun
torchrun --nproc_per_node=8 \
-m torchtitan.train \
--job.config_file ./llama3_8b_custom.toml
Step 4: Monitor and checkpoint
TensorBoard logs are saved to ./outputs/tb/:
tensorboard --logdir ./outputs/tb
Workflow 2: Multi-node training with SLURM
Multi-Node Training:
- [ ] Step 1: Configure parallelism for scale
- [ ] Step 2: Set up SLURM script
- [ ] Step 3: Submit job
- [ ] Step 4: Resume from checkpoint
Step 1: Configure parallelism for scale
For 70B model on 256 GPUs (32 nodes):
[parallelism]
data_parallel_shard_degree = 32 # FSDP across 32 ranks
tensor_parallel_degree = 8 # TP within node
pipeline_parallel_degree = 1 # No PP for 70B
context_parallel_degree = 1 # Increase for long sequences
Step 2: Set up SLURM script
#!/bin/bash
#SBATCH --job-name=llama70b
#SBATCH --nodes=32
#SBATCH --ntasks-per-node=8
#SBATCH --gpus-per-node=8
srun torchrun \
--nnodes=32 \
--nproc_per_node=8 \
--rdzv_backend=c10d \
--rdzv_endpoint=$MASTER_ADDR:$MASTER_PORT \
-m torchtitan.train \
--job.config_file ./llama3_70b.toml
Step 3: Submit job
sbatch multinode_trainer.slurm
Step 4: Resume from checkpoint
Training auto-resumes if checkpoint exists in configured folder.
Workflow 3: Enable Float8 training for H100s
Float8 provides 30-50% speedup on H100 GPUs.
Float8 Training:
- [ ] Step 1: Install torchao
- [ ] Step 2: Configure Float8
- [ ] Step 3: Launch with compile
Step 1: Install torchao
USE_CPP=0 pip install git+https://github.com/pytorch/ao.git
Step 2: Configure Float8
Add to your TOML config:
[model]
converters = ["quantize.linear.float8"]
[quantize.linear.float8]
enable_fsdp_float8_all_gather = true
precompute_float8_dynamic_scale_for_fsdp = true
filter_fqns = ["output"] # Exclude output layer
[compile]
enable = true
components = ["model", "loss"]
Step 3: Launch with compile
CONFIG_FILE="./llama3_8b.toml" ./run_train.sh \
--model.converters="quantize.linear.float8" \
--quantize.linear.float8.enable_fsdp_float8_all_gather \
--compile.enable
Workflow 4: 4D parallelism for 405B models
4D Parallelism (FSDP + TP + PP + CP):
- [ ] Step 1: Create seed checkpoint
- [ ] Step 2: Configure 4D parallelism
- [ ] Step 3: Launch on 512 GPUs
Step 1: Create seed checkpoint
Required for consistent initialization across PP stages:
NGPU=1 CONFIG_FILE=./llama3_405b.toml ./run_train.sh \
--checkpoint.enable \
--checkpoint.create_seed_checkpoint \
--parallelism.data_parallel_shard_degree 1 \
--parallelism.tensor_parallel_degree 1 \
--parallelism.pipeline_parallel_degree 1
Step 2: Configure 4D parallelism
[parallelism]
data_parallel_shard_degree = 8 # FSDP
tensor_parallel_degree = 8 # TP within node
pipeline_parallel_degree = 8 # PP across nodes
context_parallel_degree = 1 # CP for long sequences
[training]
local_batch_size = 32
seq_len = 8192
Step 3: Launch on 512 GPUs
# 64 nodes x 8 GPUs = 512 GPUs
srun torchrun --nnodes=64 --nproc_per_node=8 \
-m torchtitan.train \
--job.config_file ./llama3_405b.toml
When to use vs alternatives
Use TorchTitan when:
- Pretraining LLMs from scratch (8B to 405B+)
- Need PyTorch-native solution without third-party dependencies
- Require composable 4D parallelism (FSDP2, TP, PP, CP)
- Training on H100s with Float8 support
- Want interoperable checkpoints with torchtune/HuggingFace
Use alternatives instead:
- Megatron-LM: Maximum performance for NVIDIA-only deployments
- DeepSpeed: Broader ZeRO optimization ecosystem, inference support
- Axolotl/TRL: Fine-tuning rather than pretraining
- LitGPT: Educational, smaller-scale training
Common issues
Issue: Out of memory on large models
Enable activation checkpointing and reduce batch size:
[activation_checkpoint]
mode = "full" # Instead of "selective"
[training]
local_batch_size = 1
Or use gradient accumulation:
[training]
local_batch_size = 1
global_batch_size = 32 # Accumulates gradients
Issue: TP causes high memory with async collectives
Set environment variable:
export TORCH_NCCL_AVOID_RECORD_STREAMS=1
Issue: Float8 training not faster
Float8 only benefits large GEMMs. Filter small layers:
[quantize.linear.float8]
filter_fqns = ["attention.wk", "attention.wv", "output", "auto_filter_small_kn"]
Issue: Checkpoint loading fails after parallelism change
Use DCP's resharding capability:
# Convert sharded checkpoint to single file
python -m torch.distributed.checkpoint.format_utils \
dcp_to_torch checkpoint/step-1000 checkpoint.pt
Issue: Pipeline parallelism initialization
Create seed checkpoint first (see Workflow 4, Step 1).
Supported models
| Model | Sizes | Status |
|---|---|---|
| Llama 3.1 | 8B, 70B, 405B | Production |
| Llama 4 | Various | Experimental |
| DeepSeek V3 | 16B, 236B, 671B (MoE) | Experimental |
| GPT-OSS | 20B, 120B (MoE) | Experimental |
| Qwen 3 | Various | Experimental |
| Flux | Diffusion | Experimental |
Performance benchmarks (H100)
| Model | GPUs | Parallelism | TPS/GPU | Techniques |
|---|---|---|---|---|
| Llama 8B | 8 | FSDP | 5,762 | Baseline |
| Llama 8B | 8 | FSDP+compile+FP8 | 8,532 | +48% |
| Llama 70B | 256 | FSDP+TP+AsyncTP | 876 | 2D parallel |
| Llama 405B | 512 | FSDP+TP+PP | 128 | 3D parallel |
Advanced topics
FSDP2 configuration: See references/fsdp.md for detailed FSDP2 vs FSDP1 comparison and ZeRO equivalents.
Float8 training: See references/float8.md for tensorwise vs rowwise scaling recipes.
Checkpointing: See references/checkpoint.md for HuggingFace conversion and async checkpointing.
Adding custom models: See references/custom-models.md for TrainSpec protocol.
Resources
- GitHub: https://github.com/pytorch/torchtitan
- Paper: https://arxiv.org/abs/2410.06511
- ICLR 2025: https://iclr.cc/virtual/2025/poster/29620
- PyTorch Forum: https://discuss.pytorch.org/c/distributed/torchtitan/44
Frequently asked questions about TorchTitan
Similar skills
Python PyPI Package Builder
Streamline the process of creating and publishing Python packages.
Minecraft Plugin Development
Streamline your Minecraft server plugin creation.
MCP Server Builder
Easily build .NET MCP servers with the latest standards.
CommunityToolkit.Mvvm Messenger
Decoupled communication for ViewModels in .NET applications.
MVVM Toolkit DI
Streamline ViewModel integration with Dependency Injection in .NET.
MCP Apps Builder
Essential guidelines for MCP server development.
