test: pin custom functional launchers to FlashAttention 2

5b38ce0

Select commit

Loading

Failed to load commit list.

Uh oh!

There was an error while loading. Please reload this page.

DCO

DCO

Auto Swap Labels on: pull_request_target 2

check-approval

DCO gate pull request on: push 1

DCO gate

Request NVSkills CI on: push

request

Codeowners Approval Workflow on: push 3

pre-flight / pre-flight

codeowners-approval

multi-approval-bot-summary

Copyright check on: push 3

pre-flight / pre-flight

copyright-check / main

copyright-check-summary

Process group usage check on: push 2

pre-flight / pre-flight

process-group-usage-check

Installation Test on: push 2

pre-flight / pre-flight

Pip - Python3.12 - AMD64/Linux - NGC PyTorch

UV - Python3.12 - AMD64/Linux - NGC PyTorch

Install test summary

CICD Megatron-LM on: push 477

ephemeral-runner-routing / resolve

merge-queue-notification

DCO

is-not-external-contributor

pre-flight / pre-flight

configure

linting

cicd-wait-in-queue

cicd-parse-downstream-testing

cicd-container-build / cicd-compute-build-matrix

cicd-container-build / Build CI container (aws-h100)

cicd-container-build / Build CI container (gb-gpu)

cicd-mbridge-testing

cicd-nemo-rl-testing

cicd-parse-unit-tests

cicd-parse-unit-tests-gb200

cicd-mbridge-testing-notify

cicd-nemo-rl-testing-notify

tests/unit_tests/test_model_configs.py - latest

tests/unit_tests/test_fp8_param.py - latest

tests/unit_tests/pipeline_parallel/**/*.py - latest

tests/unit_tests/models/**/*.py - latest

tests/unit_tests/data/**/*.py - latest

tests/unit_tests/dist_checkpointing/test_optimizer.py - latest

tests/unit_tests/dist_checkpointing/test_layer_wise_optimizer_common_state_dict.py - latest

tests/unit_tests/dist_checkpointing/**/*.py - latest

tests/unit_tests/dist_checkpointing/models/**/*.py - latest

tests/unit_tests/dist_checkpointing/models/test_moe_experts.py - latest

tests/unit_tests/transformer/experimental_attention_variant/test_dsa_tilelang_kernels.py - latest

tests/unit_tests/transformer/**/*.py - latest

tests/unit_tests/inference/engines/test_dynamic_engine.py - latest

tests/unit_tests/inference/engines/test_dynamic_engine_parallel_and_chunked_prefill.py - latest

tests/unit_tests/inference/test_hybrid_moe.py - latest

tests/unit_tests/inference/**/*.py - latest

tests/unit_tests/rl/**/*.py - latest

tests/unit_tests/post_training/**/*.py - latest

tests/unit_tests/resharding/**/*.py - latest

tests/unit_tests/ssm/test_gated_delta_net.py - latest

tests/unit_tests/ssm/test_gated_delta_net_recompute.py - latest

tests/unit_tests/ssm/test_gated_delta_net_padding.py - latest

tests/unit_tests/ssm/test_gated_delta_net_parallel.py - latest

tests/unit_tests/ssm/test_gated_delta_net2_parallel.py - latest

tests/unit_tests/ssm/test_gdp_dynamic_inference.py - latest

tests/unit_tests/ssm/test_gdn_gated_output_norm_fusion.py - latest

tests/unit_tests/ssm/**/*.py - latest

tests/unit_tests/transformer/moe/test_token_dispatcher_capacity.py - latest

tests/unit_tests/transformer/moe/**/*.py - latest

tests/unit_tests/distributed/mfsdp_v1/test_mfsdp_fully_shard.py - latest

tests/unit_tests/distributed/mfsdp_v1/**/*.py - latest

tests/unit_tests/distributed/mfsdp_v2/**/*.py - latest

tests/unit_tests/a2a_overlap/**/*.py - latest

tests/unit_tests/generalized_tensor_parallel/**/*.py - latest

tests/unit_tests/determinism/correctness/**/*.py - latest

tests/unit_tests/determinism/kernels/**/*.py - latest

tests/unit_tests/**/*.py - latest

tests/unit_tests/transformer/test_multi_latent_attention.py - latest

tests/unit_tests/transformer/test_multi_token_prediction.py - latest

tests/unit_tests/**/*.py - gb200 latest

tests/unit_tests/resharding/test_nccl_m2n_copy_service.py - gb200 latest

tests/unit_tests/generalized_tensor_parallel/**/*.py - gb200 latest

cicd-integration-gate

cicd-parse-integration-tests-h100

cicd-parse-integration-tests-gb200

gpt/determinism_perf - latest

gpt/gpt3_mcore_te_mfsdp_v2_cp2 - latest

gpt/gpt3_mcore_te_tp1_pp1_dist_optimizer_fim_dataset - latest

gpt/gpt3_mcore_te_tp1_pp1_dist_optimizer_no_mmap_bin_files - latest

gpt/gpt3_mcore_te_tp1_pp1_resume_torch_dist_uniform_full_recompute - latest

gpt/gpt3_mcore_te_tp1_pp2_resume_torch_dist_rope_embeddings - latest

gpt/gpt3_mcore_te_tp1_pp2_resume_torch_dist_rope_embeddings_interleaved_no_fusion - latest

gpt/gpt3_mcore_te_tp1_pp4_resume_torch_dist_disable_bias_linear - latest

gpt/gpt3_mcore_te_tp1_pp4_resume_torch_dist_persistent_disable_bias_linear - latest

gpt/gpt3_mcore_te_tp1_pp4_resume_torch_dist_untie_embeddings_and_outputs - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1 - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_param_gather_overlap_optimizer - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_resume_torch_dist_calculate_per_token_loss - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_resume_torch_dist_dist_optimizer_overlap_grad_reduce - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_uneven_pipeline - latest

gpt/gpt3_mcore_te_tp1_pp4_vp2_account_for_embedding_loss_in_pipeline_split - latest

gpt/gpt3_mcore_te_tp2_pp1_gdn - latest

gpt/gpt3_mcore_te_tp2_pp1_gdn_no_nvrx_async - latest

gpt/gpt3_mcore_te_tp2_pp1_gdn_no_nvrx_sync - latest

gpt/gpt3_mcore_te_tp2_pp1_modelopt_distill_resume - latest

gpt/gpt3_mcore_te_tp2_pp1_resume_torch_dist_cp2_nondeterministic - latest

gpt/gpt3_mcore_te_tp2_pp1_resume_torch_dist_multi_dist_optimizer_instances - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2 - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_calculate_per_token_loss - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_calculate_per_token_loss_nondeterministic - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_etp4_calculate_per_token_loss_dp_last - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_etp4_calculate_per_token_loss_nondeterministic_dp_last - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_etp4_dp_last - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_etp4_nondeterministic_dp_last - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_nondeterministic - latest

gpt/gpt3_mcore_te_tp2_pp2_cross_entropy_loss_fusion - latest

gpt/gpt3_mcore_te_tp2_pp2_dsa - latest

gpt/gpt3_mcore_te_tp2_pp2_mla - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_cp2_nondeterministic - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_cross_entropy_loss_fusion - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_ddp_average_in_collective - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_defer_embedding_wgrad_compute - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_reshard_1x4xNone - latest

gpt/gpt3_mcore_te_tp4_pp1_dist_optimizer_overlap_grad_reduce_param_gather - latest

gpt/gpt3_mcore_te_tp4_pp1_resume_torch_dist_dist_optimizer_overlap_grad_reduce - latest

gpt/gpt3_mcore_te_tp4_pp1_resume_torch_dist_dist_optimizer_overlap_grad_reduce_param_gather - latest

gpt/gpt3_mcore_te_tp4_pp1_resume_torch_dist_qk_layernorm_test_mode - latest

gpt/gpt3_mcore_te_tp4_pp2_resume_torch_dist_reshard_8x1xNone - latest

gpt/gpt3_mcore_tp2_pp2_uninstall_te - latest

gpt/gpt_dynamic_inference_tp1_pp1_dp8_583m_long_prompt_logitsmatch_zmq - latest

gpt/gpt_dynamic_inference_tp2_pp2_dp2_583m_logitsmatch_zmq - latest

gpt/gpt_grpo_basic_function - latest

gpt/gpt_inference_server_async_sched_tp2_pp2_dp2_qwen3_8b_sse - latest

gpt/gpt_static_inference_tp1_pp1_16b_multiprompt_tokensmatch - latest

gpt/gpt_static_inference_tp1_pp1_583m_cudagraphs - latest

gpt/gpt_static_inference_tp1_pp1_583m_fp8_cudagraphs - latest

gpt/gpt_static_inference_tp1_pp1_583m_logitsmatch - latest

hybrid/hybrid_dynamic_inference_tp1_pp1_dp8_583m - latest

hybrid/hybrid_dynamic_inference_tp1_pp1_dp8_583m_chunked_prefill - latest

hybrid/hybrid_dynamic_inference_tp1_pp1_dp8_583m_flashinfer - latest

hybrid/hybrid_dynamic_inference_tp1_pp1_dp8_583m_mamba_bf16_states - latest

hybrid/hybrid_gdp_dynamic_inference_tp1_pp1_dp8_800m - latest

hybrid/hybrid_llama3_proxy_mfsdp_v2_cuda_graph - latest

hybrid/hybrid_mr_mcore_te_tp1_pp1_cp1_dgx_a100_1N8G - latest

hybrid/hybrid_mr_mcore_te_tp1_pp2_vpp2_cp1_dgx_a100_1N8G - latest

hybrid/hybrid_mr_mcore_te_tp1_pp4_cp1_dgx_a100_1N8G - latest

hybrid/hybrid_mr_mcore_te_tp2_pp1_cp1_dgx_a100_1N8G - latest

hybrid/hybrid_mr_mcore_te_tp2_pp1_cp4_dgx_a100_1N8G - latest

mimo/mimo_hetero_gtp_checkpoint_round_trip - latest

moe/deepseek_proxy_mfsdp_v2_ep2 - latest

moe/deepseek_proxy_mfsdp_v2_ep2_ep_overlap - latest

moe/deepseek_proxy_mfsdp_v2_ep2_hfsdp_mb2 - latest

moe/deepseek_proxy_mfsdp_v2_ep2_hsdp_mb2 - latest

moe/deepseek_proxy_mfsdp_v2_ep2_mb2 - latest

moe/gpt3_mcore_te_tp1_pp1_te_4experts_groupedGEMM_op_fuser - latest

moe/gpt3_mcore_te_tp1_pp2_resume_torch_dist_reshard_2x1x4_te_8experts2parallel_dist_optimizer - latest

moe/gpt3_mcore_te_tp2_pp1_resume_torch_dist_te_8experts2parallel_dist_optimizer - latest

moe/gpt3_mcore_te_tp2_pp1_resume_torch_dist_te_8experts2parallel_multi_dist_optimizer_instances - latest

moe/gpt3_mcore_te_tp2_pp1_te_8experts2parallel_ddp_average_in_collective - latest

moe/gpt3_mcore_te_tp2_pp1_te_8experts2parallel_overlap_grad_reduce_param_gather_groupedGEMM - latest

moe/gpt3_mcore_te_tp2_pp1_te_a2a_ovlp_8experts_etp1_ep4 - latest

moe/gpt3_moe_mcore_te_tp4_ep2_etp2_pp2_resume_torch_dist_dist_optimizer - latest

moe/gpt3_moe_mcore_te_tp4_ep2_etp2_pp2_scoped_cudagraph - latest

moe/gpt_dynamic_inference_tp4_pp1_ep4_16B_mtp_sp_block_cudagraphs - latest

gpt/gpt3_mcore_reruns_persistent_1_1node - latest

gpt/gpt3_mcore_te_tp1_pp1_dist_optimizer_no_mmap_bin_files_1node - latest

gpt/gpt3_mcore_te_tp1_pp1_resume_torch_dist_dist_optimizer_1node - latest

gpt/gpt3_mcore_te_tp1_pp2_resume_torch_dist_rope_embeddings_interleaved_no_fusion_1node - latest

gpt/gpt3_mcore_te_tp1_pp4_resume_torch_dist_persistent_disable_bias_linear_1node - latest

gpt/gpt3_mcore_te_tp1_pp4_resume_torch_dist_untie_embeddings_and_outputs_1node - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_resume_torch_decoupled_lr_1node - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_resume_torch_dist_dist_optimizer_overlap_grad_reduce_1node - latest

gpt/gpt3_mcore_te_tp1_pp4_vp1_uneven_pipeline_1node - latest

gpt/gpt3_mcore_te_tp2_pp1_resume_torch_dist_cp2_nondeterministic_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_calculate_per_token_loss_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_calculate_per_token_loss_nondeterministic_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_cp2_nondeterministic_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_cross_entropy_loss_fusion_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_mla_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_cp2_nondeterministic_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_defer_embedding_wgrad_compute_1node - latest

gpt/gpt3_mcore_te_tp2_pp2_resume_torch_dist_reshard_1x4xNone_1node - latest

gpt/gpt3_mcore_te_tp4_pp1_resume_torch_dist_dist_optimizer_overlap_grad_reduce_param_gather_1node - latest

gpt/gpt3_mcore_te_tp4_pp1_resume_torch_dist_qk_layernorm_test_mode_1node - latest

gpt/gpt3_mcore_te_tp4_pp2_resume_torch_dist_reshard_8x1xNone_1node - latest

hybrid/hybrid_gdp_dynamic_inference_tp1_pp1_dp4_800m - latest

moe/gpt3_mcore_te_tp2_pp1_resume_torch_dist_te_8experts2parallel_dist_optimizer_1node - latest

moe/gpt3_mcore_te_tp2_pp1_te_8experts2parallel_ddp_average_in_collective_1node - latest

moe/gpt3_mcore_te_tp2_pp1_te_8experts2parallel_overlap_grad_reduce_param_gather_groupedGEMM_1node - latest

moe/gpt3_moe_mcore_te_ep8_resume_torch_dist_muon_1node - latest

moe/gpt3_moe_mcore_te_tp4_ep2_etp2_pp2_resume_torch_dist_dist_optimizer_1node - latest

moe/gpt_dynamic_inference_tp4_pp1_ep4_16B_inference_optimized_swiglu_logitsmatch - latest

nemotron/nemotron3_5_lightning_no_load_optim_tp1_pp1_cp1_ep4_dgx_gb200_1N4G - latest

Nemo_CICD_Test

Coverage_Fake

Coverage (unit-test)

Build, validate, and release Megatron-Core on: push 20

validate-inputs

pre-flight / pre-flight

bump / check-admin-permission

bump / bump-next-version

build-test-publish-wheels / build-and-test-wheels (megatron-core, arm64, quay.io/pypa/manylinux_2_28_aarch64)

build-test-publish-wheels / build-and-test-wheels (megatron-core, amd64, quay.io/pypa/manylinux_2_28_x86_64)

build-test-publish-wheels / build-and-test-wheels (megatron-fsdp, quay.io/pypa/manylinux_2_28_x86_64, amd64)

build-test-publish-wheels / publish-wheels (megatron-core, arm64)

build-test-publish-wheels / publish-wheels (megatron-core, amd64)

build-test-publish-wheels / publish-wheels (megatron-fsdp, amd64)

finalize / create-gh-release

finalize / build-docs / Build docs

finalize / notify

finalize / publish-docs

release-summary

Review Trigger on: pull_request_review

signal

Review Trigger on: pull_request_review

signal

Review Trigger on: pull_request_review 1

signal

Review Trigger on: pull_request_review 1

signal

Review Trigger on: pull_request_review 1

signal

Review Trigger on: pull_request_review 1

signal

DCO / DCO succeeded Oct 8, 2026 in 1s

DCO

All commits are signed off!

View more details on DCO

Loading

Re-running checks...