MLM vs Bridge Training#
For how they differ, the arg mapping tables, gotchas, and translation script, see:
docs/megatron-lm-to-megatron-bridge.md
Correlation Testing#
Use vanilla_gpt_pretrain_config for loss-correlation testing. This recipe uses
bare GPTModelProvider defaults (LayerNorm, GeLU, learned_absolute position
embeddings, vocab_size inherited from tokenizer) — matching MLM
pretrain_gpt.py defaults with no args.
MLM Correlation Run (2L/256H, 1 GPU)#
PYTHONPATH=3rdparty/Megatron-LM:$PYTHONPATH \
uv run python -m torch.distributed.run --nproc_per_node=1 \
3rdparty/Megatron-LM/pretrain_gpt.py \
--num-layers 2 --hidden-size 256 --num-attention-heads 4 \
--ffn-hidden-size 1024 --seq-length 512 --max-position-embeddings 512 \
--micro-batch-size 4 --global-batch-size 32 \
--train-iters 10 --eval-iters 2 --eval-interval 10 \
--mock-data --bf16 --use-mcore-models \
--tokenizer-type NullTokenizer --vocab-size 32000 \
--lr 3e-4 --min-lr 3e-5 --seed 1234 --log-interval 1
Bridge Correlation Run (same config, 1 GPU)#
rm -rf nemo_experiments && \
uv run python -m torch.distributed.run --nproc_per_node=1 \
scripts/training/run_recipe.py \
--recipe vanilla_gpt_pretrain_config \
model.num_layers=2 model.hidden_size=256 \
model.num_attention_heads=4 model.ffn_hidden_size=1024 \
model.seq_length=512 dataset.sequence_length=512 \
train.train_iters=10 train.global_batch_size=32 train.micro_batch_size=4 \
validation.eval_interval=10 validation.eval_iters=2 \
optimizer.lr=3e-4 optimizer.min_lr=3e-5 \
scheduler.lr_warmup_iters=1 scheduler.lr_decay_iters=10 \
rng.seed=1234 logger.log_interval=1
Verification#
With matched parameters the LM losses should be nearly identical at each
iteration. Compare lm loss values from both logs — they should agree to
within BF16 rounding.
Multi-GPU Examples#
MLM 2-GPU with TP=2#
PYTHONPATH=3rdparty/Megatron-LM:$PYTHONPATH \
uv run python -m torch.distributed.run --nproc_per_node=2 \
3rdparty/Megatron-LM/pretrain_gpt.py \
--tensor-model-parallel-size 2 --sequence-parallel \
--num-layers 4 --hidden-size 256 --num-attention-heads 4 \
--seq-length 1024 --max-position-embeddings 1024 \
--micro-batch-size 2 --global-batch-size 16 \
--train-iters 10 --eval-iters 2 --eval-interval 10 \
--mock-data --bf16 --use-mcore-models \
--tokenizer-type NullTokenizer --vocab-size 1024 \
--lr 1e-4 --log-interval 1
Bridge 2-GPU with TP=2#
rm -rf nemo_experiments && \
uv run python -m torch.distributed.run --nproc_per_node=2 \
scripts/training/run_recipe.py \
--recipe vanilla_gpt_pretrain_config \
model.tensor_model_parallel_size=2 model.sequence_parallel=true \
model.num_layers=4 model.hidden_size=256 \
model.num_attention_heads=4 model.ffn_hidden_size=1024 \
model.seq_length=1024 dataset.sequence_length=1024 \
train.train_iters=10 train.global_batch_size=16 train.micro_batch_size=2 \
validation.eval_interval=10 validation.eval_iters=2 \
scheduler.lr_warmup_iters=2 scheduler.lr_decay_iters=10 \
logger.log_interval=1
Available Recipes#
Common recipes (use with --recipe):
vanilla_gpt_pretrain_config— Minimal GPT (bare GPTModelProvider defaults, ideal for correlation testing and custom configs)llama32_1b_pretrain_config— Llama 3.2 1B (16L, 2048H, GBS=512, seq=8192)llama3_8b_pretrain_config— Llama 3 8Bqwen3_8b_pretrain_config— Qwen3 8Bdeepseek_v2_lite_pretrain_config— DeepSeek-V2-Lite 16B MoE
SFT/PEFT variants use _sft_config / _peft_config suffix.
Megatron-Core Submodule#
For what the submodule is and why two versions exist, see
docs/megatron-lm-to-megatron-bridge.md.
Check current version#
./scripts/switch_mcore.sh status
Switch to dev for testing newer MCore features#
./scripts/switch_mcore.sh dev
# uv sync (without --locked) since lockfile is for main
uv sync
Switch back to main#
./scripts/switch_mcore.sh main
After pulling latest main#
When you pull the latest Bridge main branch, the submodule pointer may have been updated. Re-sync the submodule:
git submodule update --init 3rdparty/Megatron-LM
Pitfalls#
Always
rm -rf nemo_experimentsbefore a fresh correlation run. Bridge auto-resumes from stale checkpoints silently.uv runrequired: Always useuv run python -m torch.distributed.run(not baretorchrunorpython).MLM PYTHONPATH: Must include
3rdparty/Megatron-LMsogpt_builders.pyis importable.Scheduler overrides: When overriding
train.train_itersto a small value, also setscheduler.lr_warmup_itersandscheduler.lr_decay_itersor you get an assertion error.Use
dataset.sequence_lengthin CLI overrides, notdataset.seq_length.MoE OOM: Large MoE models require full activation recomputation and typically multi-node EP. TP does NOT reduce per-GPU expert memory.
uv sync --lockedfails after switching to dev: The lockfile is generated against the main MCore commit. Useuv sync(without--locked) when on dev.