Skip to content

chore: AUT-2517 bump MCore main to fde43b6d #6405

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ko3n1g merged 2 commits into from
Oct 11, 2026
Merged

chore: AUT-2517 bump MCore main to fde43b6d #6405

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
67230b4
chore: bump Megatron-Core main to fde43b6d
svcnemo-autobot Oct 9, 2026
fcdb71f
test(qwen3-vl): set attention_cp_layout on process_mtp_loss config stub
svcnemo-autobot Oct 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .main.commit
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1 @@
8949def81eb744b9219e2d794c80265928c40a5b
fde43b6d617752433310b4668c6b5c55e503521a
2 changes: 1 addition & 1 deletion 3rdparty/Megatron-LM
Open in desktop
Submodule Megatron-LM updated 65 files
+11 −0 .github/workflows/cicd-main.yml
+4 −0 .github/workflows/process-group-usage-check.yml
+3 −0 docker/Dockerfile.ci.dev
+37 −27 examples/rl/benchmark_refit.py
+1 −1 examples/rl/model_configs/common.sh
+143 −0 megatron/core/context_parallel/sequence_roll.py
+19 −3 megatron/core/distributed/fsdp/src/megatron_fsdp/experimental/layout.py
+10 −10 megatron/core/distributed/fsdp/src/megatron_fsdp/experimental/owner_planning.py
+8 −14 megatron/core/distributed/fsdp/src/megatron_fsdp/experimental/quantized_dbuffer.py
+94 −0 megatron/core/fusions/fused_row_copy.py
+2 −1 megatron/core/inference/contexts/dynamic_context.py
+1 −2 megatron/core/inference/engines/dynamic_engine.py
+2 −2 megatron/core/inference/text_generation_controllers/mtp_controller_mixin.py
+6 −12 megatron/core/inference/text_generation_controllers/text_generation_controller.py
+11 −0 megatron/core/inference/utils.py
+1 −1 megatron/core/models/common/utils.py
+1 −1 megatron/core/models/gpt/fine_grained_callables.py
+95 −0 megatron/core/tensor_parallel/mappings.py
+509 −2 megatron/core/transformer/experimental_attention_variant/csa.py
+372 −0 megatron/core/transformer/experimental_attention_variant/csa_utils/cp_utils.py
+72 −0 megatron/core/transformer/experimental_attention_variant/csa_utils/csa_indexer_loss_kernels.py
+199 −35 megatron/core/transformer/experimental_attention_variant/csa_utils/csa_teacher_lse.py
+252 −0 megatron/core/transformer/experimental_attention_variant/csa_utils/packed_layout.py
+727 −0 megatron/core/transformer/experimental_attention_variant/csa_utils/packed_sparse_attention.py
+346 −115 megatron/core/transformer/experimental_attention_variant/deepseek_v4_hybrid_attention.py
+1 −0 megatron/core/transformer/moe/README.md
+232 −1 megatron/core/transformer/moe/fused_a2a.py
+2 −2 megatron/core/transformer/moe/moe_utils.py
+124 −1 megatron/core/transformer/moe/token_dispatcher.py
+17 −2 megatron/core/transformer/multi_latent_attention.py
+125 −70 megatron/core/transformer/multi_token_prediction.py
+20 −12 megatron/core/transformer/transformer_config.py
+1 −8 megatron/inference/utils.py
+58 −40 megatron/rl/inference/megatron.py
+53 −46 megatron/rl/rl_utils.py
+26 −85 megatron/training/arguments.py
+5 −2 megatron/training/checkpointing.py
+15 −0 megatron/training/config/container.py
+51 −18 megatron/training/config/inference_config.py
+55 −15 megatron/training/config/rl_config.py
+67 −59 megatron/training/training.py
+2 −2 tests/functional_tests/dynamic_inference_functional_tests.md
+2 −2 .../test_cases/moe/gpt_dynamic_inference_tp4_etp1_pp1_ep8_16B_logitsmatch_zmq_suspend_resume/model_config.yaml
+29 −1 tests/unit_tests/determinism/kernels/manifest.py
+100 −0 tests/unit_tests/determinism/kernels/test_csa_packed_kernels.py
+95 −1 tests/unit_tests/determinism/kernels/test_fused_triton_kernels.py
+22 −4 tests/unit_tests/determinism/kernels/test_moe_kernels.py
+36 −32 tests/unit_tests/distributed/mfsdp_v2/test_owner_planning.py
+42 −0 tests/unit_tests/fusions/test_fused_row_copy.py
+102 −21 tests/unit_tests/inference/test_inference_config.py
+24 −1 tests/unit_tests/inference/test_inference_utils.py
+1 −0 tests/unit_tests/inference/text_generation_controllers/test_mtp_kv_cache_controller.py
+24 −11 tests/unit_tests/rl/test_rl_utils.py
+1 −1 tests/unit_tests/rl/test_rollout_generation.py
+0 −12 tests/unit_tests/test_check_process_group_usage.py
+0 −1 tests/unit_tests/test_eval_batch_size.py
+2 −3 tests/unit_tests/test_training.py
+104 −8 tests/unit_tests/training/config/test_rl_config.py
+5 −3 tests/unit_tests/transformer/experimental_attention_variant/test_dsv4_hybrid_attention.py
+346 −0 tests/unit_tests/transformer/experimental_attention_variant/test_dsv4_packed_cp.py
+13 −8 tests/unit_tests/transformer/moe/test_grouped_tensor_dispatcher_numerics.py
+13 −3 tests/unit_tests/transformer/moe/test_token_dispatcher.py
+67 −0 tests/unit_tests/transformer/test_contiguous_sequence_roll.py
+64 −0 tests/unit_tests/transformer/test_multi_latent_attention.py
+12 −12 train_rl.py
2 changes: 2 additions & 0 deletions tests/unit_tests/models/qwen_vl/modelling_qwen3_vl/test_text_model_forward.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -254,6 +254,8 @@ def fake_output_layer(hidden, weight=None, runtime_gather_output=None): # noqa:
mtp_loss_scaling_factor=1.0,
calculate_per_token_loss=True,
use_mup=False,
# process_mtp_loss reads the CP layout; mirror TransformerConfig's default.
attention_cp_layout="zigzag",
)
output = process_mtp_loss(
hidden_states=hidden_states,
Expand Down
Loading
Loading