Training a model with Primus and Megatron-Core

Training a model with Primus and Megatron-Core#

2025-08-22

13 min read time

Applies to Linux and Windows

Primus is a unified and flexible LLM training framework designed to streamline training. It streamlines LLM training on AMD Instinct accelerators using a modular, reproducible configuration paradigm. Primus is backend-agnostic and supports multiple training engines – including Megatron-Core.

Note

Primus with the Megatron-Core backend is intended to replace ROCm Megatron-LM in this Dockerized training environment. To learn how to migrate workloads from Megatron-LM to Primus with Megatron-Core, see Migrating workloads to Primus (Megatron-Core backend) from Megatron-LM.

For ease of use, AMD provides a ready-to-use Docker image for MI300 series accelerators containing essential components for Primus and Megatron-Core.

Note

This Docker environment is based on Python 3.10 and Ubuntu 22.04. For an alternative environment with Python 3.12 and Ubuntu 24.04, see the previous ROCm Megatron-LM v25.6 Docker release.

Software component

Version

ROCm

6.4.2

Primus

v0.1.0-rc1

PyTorch

2.8.0a0+gitd06a406

Python

3.10

Transformer Engine

2.1.0.dev0+ba586519

hipBLASLt

37ba1d36

Triton

3.3.0

RCCL

2.22.3

Supported models#

The following models are pre-optimized for performance on AMD Instinct MI300X series accelerators. Some instructions, commands, and training examples in this documentation might vary by model – select one to get started.

Model
Meta Llama
DeepSeek
Mistral AI
Qwen
Model variant
Llama 3.3 70B
Llama 3.1 70B
Llama 3.1 8B
Llama 2 7B
Llama 2 70B
DeepSeek-V3 (proxy)
DeepSeek-V2-Lite
Mixtral 8x7B
Mixtral 8x22B (proxy)
Qwen 2.5 7B
Qwen 2.5 72B

Note

Some models, such as Llama, require an external license agreement through a third party (for example, Meta).

System validation#

Before running AI workloads, it’s important to validate that your AMD hardware is configured correctly and performing optimally.

If you have already validated your system settings, including aspects like NUMA auto-balancing, you can skip this step. Otherwise, complete the procedures in the System validation and optimization guide to properly configure your system settings before starting training.

To test for optimal performance, consult the recommended System health benchmarks. This suite of tests will help you verify and fine-tune your system’s configuration.

Environment setup#

Use the following instructions to set up the environment, configure the script to train models, and reproduce the benchmark results on MI300X series accelerators with the rocm/megatron-lm:v25.7_py310 image.

Download the Docker image#

  1. Use the following command to pull the Docker image from Docker Hub.

    docker pull rocm/megatron-lm:v25.7_py310

  2. Launch the Docker container.

    docker run -it \
    --device /dev/dri \
    --device /dev/kfd \
    --device /dev/infiniband \
    --network host --ipc host \
    --group-add video \
    --cap-add SYS_PTRACE \
    --security-opt seccomp=unconfined \
    --privileged \
    -v $HOME:$HOME \
    --shm-size 128G \
    --name primus_training_env \
    rocm/megatron-lm:v25.7_py310

  1. Use these commands if you exit the primus_training_env container and need to return to it.

    docker start primus_training_env
docker exec -it primus_training_env bash

The Docker container hosts verified release tag v0.1.0-rc1 of the Primus repository.

Configuration#

Primus defines a training configuration in YAML for each model in examples/megatron/configs.

To update training parameters for Llama 3.3 70B, you can update examples/megatron/configs/llama3.3_70B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Llama 3.1 70B, you can update examples/megatron/configs/llama3.1_70B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Llama 3.1 8B, you can update examples/megatron/configs/llama3.1_8B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Llama 2 7B, you can update examples/megatron/configs/llama2_7B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Llama 2 70B, you can update examples/megatron/configs/llama2_70B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for DeepSeek-V3 (proxy), you can update examples/megatron/configs/deepseek_v3-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for DeepSeek-V2-Lite, you can update examples/megatron/configs/deepseek_v2_lite-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Mixtral 8x7B, you can update examples/megatron/configs/mixtral_8x7B_v0.1-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Mixtral 8x22B (proxy), you can update examples/megatron/configs/mixtral_8x22B_v0.1-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Qwen 2.5 7B, you can update examples/megatron/configs/primus_qwen2.5_7B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

To update training parameters for Qwen 2.5 72B, you can update examples/megatron/configs/qwen2.5_72B-pretrain.yaml. Note that training configuration YAML files for other models follow this naming convention.

Note

See Key options for more information on configuration options.

Dataset options#

You can use either mock data or real data for training.

  • Mock data can be useful for testing and validation. Use the mock_data field to toggle between mock and real data. The default value is true for enabled.

    mock_data: true

  • If you’re using a real dataset, update the train_data_path field to point to the location of your dataset.

    mock_data: false
train_data_path: /path/to/your/dataset

    Ensure that the files are accessible inside the Docker container.

Tokenizer#

In Primus, each model uses a tokenizer from Hugging Face. For example, Llama 3.1 8B model uses tokenizer_model: meta-llama/Llama-3.1-8B and tokenizer_type: Llama3Tokenizer defined in the llama3.1-8B model definition. As such, you need to set the HF_TOKEN environment variable with right permissions to access the tokenizer for each model.

# Export your HF_TOKEN in the workspace
export HF_TOKEN=<your_hftoken>

Run training#

Use the following example commands to set up the environment, configure key options, and run training on MI300X series accelerators with the AMD Megatron-LM environment.

Single node training#

To run training on a single node, navigate to /workspace/Primus and use the following setup command:

pip install -r requirements.txt
export HSA_NO_SCRATCH_RECLAIM=1
export NVTE_CK_USES_BWD_V3=1

Once setup is complete, run the appropriate training command.

To run pre-training for Llama 3.3 70B BF16, run:

EXP=examples/megatron/configs/llama3.3_70B-pretrain.yaml \
bash ./examples/run_pretrain.sh \
    --micro_batch_size 2 \
    --global_batch_size 16 \
    --train_iters 50

To run pre-training for Llama 3.1 8B FP8, run:

EXP=examples/megatron/configs/llama3.1_8B-pretrain.yaml \
bash ./examples/run_pretrain.sh \
    --train_iters 50 \
    --fp8 hybrid

For Llama 3.1 8B BF16, use the following command:

EXP=examples/megatron/configs/llama3.1_8B-pretrain.yaml \
bash ./examples/run_pretrain.sh --train_iters 50

To run pre-training for Llama 3.1 70B BF16, run:

EXP=examples/megatron/configs/llama3.1_70B-pretrain.yaml \
bash ./examples/run_pretrain.sh \
     --train_iters 50

To run the training on a single node for Llama 3.1 70B FP8 with proxy, use the following command:

EXP=examples/megatron/configs/llama3.1_70B-pretrain.yaml \
bash ./examples/run_pretrain.sh \
    --train_iters 50 \
    --num_layers 40 \
    --fp8 hybrid \
    --no_fp8_weight_transpose_cache true

Note

Use two or more nodes to run the full Llama 70B model with FP8 precision.

To run pre-training for Llama 2 7B FP8, run:

EXP=examples/megatron/configs/llama2_7B-pretrain.yaml \
bash ./examples/run_pretrain.sh \
    --train_iters 50 \
    --fp8 hybrid

To run pre-training for Llama 2 7B BF16, run:

EXP=examples/megatron/configs/llama2_7B-pretrain.yaml \
bash ./examples/run_pretrain.sh --train_iters 50

To run pre-training for Llama 2 70B BF16, run:

EXP=examples/megatron/configs/llama2_70B-pretrain.yaml \
bash ./examples/run_pretrain.sh --train_iters 50

To run training on a single node for DeepSeek-V3 (MoE with expert parallel) with 3-layer proxy, use the following command:

EXP=examples/megatron/configs/deepseek_v3-pretrain.yaml \
bash examples/run_pretrain.sh \
    --num_layers 3 \
    --moe_layer_freq 1 \
    --train_iters 50

To run training on a single node for DeepSeek-V2-Lite (MoE with expert parallel), use the following command:

EXP=examples/megatron/configs/deepseek_v2_lite-pretrain.yaml \
bash examples/run_pretrain.sh \
    --global_batch_size 256 \
    --train_iters 50

To run training on a single node for Mixtral 8x7B (MoE with expert parallel), use the following command:

EXP=examples/megatron/configs/mixtral_8x7B_v0.1-pretrain.yaml \
bash examples/run_pretrain.sh --train_iters 50

To run training on a single node for Mixtral 8x7B (MoE with expert parallel) with 4-layer proxy, use the following command:

EXP=examples/megatron/configs/mixtral_8x22B_v0.1-pretrain.yaml \
bash examples/run_pretrain.sh \
    --num_layers 4 \
    --pipeline_model_parallel_size 1 \
    --micro_batch_size 1 \
    --global_batch_size 16 \
    --train_iters 50

To run training on a single node for Qwen 2.5 7B BF16, use the following command:

EXP=examples/megatron/configs/qwen2.5_7B-pretrain.yaml \
bash examples/run_pretrain.sh --train_iters 50

For FP8, use the following command.

EXP=examples/megatron/configs/qwen2.5_7B-pretrain.yaml \
bash examples/run_pretrain.sh \
    --train_iters 50 \
    --fp8 hybrid

To run the training on a single node for Qwen 2.5 72B BF16, use the following command.

EXP=examples/megatron/configs/qwen2.5_72B-pretrain.yaml \
bash examples/run_pretrain.sh --train_iters 50

Multi-node training examples#

To run training on multiple nodes, you can use the run_slurm_pretrain.sh to launch the multi-node workload. Use the following steps to setup your environment:

cd /workspace/Primus/
export DOCKER_IMAGE=rocm/megatron-lm:v25.7_py310
export HF_TOKEN=<your_HF_token>
export HSA_NO_SCRATCH_RECLAIM=1
export NVTE_CK_USES_BWD_V3=1
export NCCL_IB_HCA=<your_NCCL_IB_HCA> # specify which RDMA interfaces to use for communication
export NCCL_SOCKET_IFNAME=<your_NCCL_SOCKET_IFNAME> # your Network Interface
export GLOO_SOCKET_IFNAME=<your_GLOO_SOCKET_IFNAME> # your Network Interface
export NCCL_IB_GID_INDEX=3 # Set InfiniBand GID index for NCCL communication. Default is 3 for ROCE

Note

  • Make sure correct network drivers are installed on the nodes. If inside a Docker, either install the drivers inside the Docker container or pass the network drivers from the host while creating Docker container.

  • If NCCL_IB_HCA and NCCL_SOCKET_IFNAME are not set, Primus will try to auto-detect. However, since NICs can vary accross different cluster, it is encouraged to explicitly export your NCCL parameters for the cluster.

  • To find your network interface, you can use ip a.

  • To find RDMA interfaces, you can use ibv_devices to get the list of all the RDMA/IB devices.

To train Llama 3.3 70B FP8 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama3.3_70B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 4 \
    --global_batch_size 256 \
    --recompute_num_layers 80 \
    --no_fp8_weight_transpose_cache true \
    --fp8 hybrid

To train Llama 3.3 70B BF16 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama3.3_70B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 1 \
    --global_batch_size 256 \
    --recompute_num_layers 12

To train Llama 3.1 8B FP8 on 8 nodes, run:

# Adjust the training parameters. For e.g., `global_batch_size: 8 * #single_node_bs` for 8 nodes in this case
NNODES=8 EXP=examples/megatron/configs/llama3.1_8B-pretrain.yaml \
bash ./examples/run_slurm_pretrain.sh \
    --global_batch_size 1024 \
    --fp8 hybrid

To train Llama 3.1 70B FP8 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama3.1_70B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 4 \
    --global_batch_size 256 \
    --recompute_num_layers 80 \
    --no_fp8_weight_transpose_cache true \
    --fp8 hybrid

To train Llama 3.1 70B BF16 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama3.1_70B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 1 \
    --global_batch_size 256 \
    --recompute_num_layers 12

To train Llama 2 8B FP8 on 8 nodes, run:

# Adjust the training parameters. For e.g., `global_batch_size: 8 * #single_node_bs` for 8 nodes in this case
NNODES=8 EXP=examples/megatron/configs/llama2_7B-pretrain.yaml bash ./examples/run_slurm_pretrain.sh --global_batch_size 2048 --fp8 hybrid

To train Llama 2 70B FP8 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama2_70B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 10 \
    --global_batch_size 640 \
    --recompute_num_layers 80 \
    --no_fp8_weight_transpose_cache true \
    --fp8 hybrid

To train Llama 2 70B BF16 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/llama2_70B-pretrain.yaml \
bash ./examples/run_slurm_pretrain.sh \
    --micro_batch_size 2 \
    --global_batch_size 1536 \
    --recompute_num_layers 12

To train Mixtral 8x7B BF16 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/mixtral_8x7B_v0.1-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 2 \
    --global_batch_size 256

To train Qwen2.5 72B FP8 on 8 nodes, run:

NNODES=8 EXP=examples/megatron/configs/qwen2.5_72B-pretrain.yaml \
bash examples/run_slurm_pretrain.sh \
    --micro_batch_size 8 \
    --global_batch_size 512 \
    --recompute_num_layers 80 \
    --no_fp8_weight_transpose_cache true \
    --fp8 hybrid

Key options#

The following are key options to take note of

fp8

hybrid enables FP8 GEMMs.

use_torch_fsdp2

use_torch_fsdp2: 1 enables torch fsdp-v2. If FSDP is enabled, set use_distributed_optimizer and overlap_param_gather to false.

profile

To enable PyTorch profiling, set these parameters:

profile: true
use_pytorch_profiler: true
profile_step_end: 7
profile_step_start: 6
train_iters

The total number of iterations (default: 50).

mock_data

True by default.

micro_batch_size

Micro batch size.

global_batch_size

Global batch size.

recompute_granularity

For activation checkpointing.

num_layers

For using a reduced number of layers as with proxy models.

Previous versions#

See Megatron-LM training performance testing version history to find documentation for previous releases of the ROCm/megatron-lm Docker image.

This training environment now uses Primus with Megatron as the primary configuration. Limited support for the legacy ROCm Megatron-LM is still available. For instructions on using ROCm Megatron-LM, see the Training a model with Megatron-LM for ROCm document.