Jiajun Li a28751c137 test(e2e): cover distributed Muon for three MoE models (#3878)

2026-10-01 15:27:50 -07:00

..

backends

feat(fsdp): load module-level compute kernels from the Hub (#2829)

2026-09-30 19:24:47 -07:00

charts

Give every worker that watches pods the platform read access it needs (#3056)

2026-09-26 20:36:27 +08:00

dashboard

Teach the dashboard and the CI checkers the policy namespace (#2649)

2026-09-26 19:10:10 +08:00

doc

Deliver the injection plan through a file the relaunch can rewrite (#3013)

2026-09-26 20:14:40 +08:00

e2e

test(e2e): cover distributed Muon for three MoE models (#3878)

2026-10-01 15:27:50 -07:00

examples

Pause generation with retract in the solver-verifier recipe (#3075)

2026-09-26 20:45:20 +08:00

fixtures

Drop the empty-batch timeout only multi-LoRA v1 ever raised (#3723)

2026-09-26 20:50:59 +08:00

launch_scripts

perf(dsv4): balance the CSA indexer across contiguous CP ranks (#3689)

2026-09-29 09:05:40 -07:00

miles_plugins/models

perf(dsv4): balance the CSA indexer across contiguous CP ranks (#3689)

2026-09-29 09:05:40 -07:00

models

…

optimizers

…

ray

fix(ray): pin each trainer to the NUMA node of the GPU it actually uses (#3869)

2026-09-29 21:07:58 -07:00

rollout

Report the raw reward of every generated group, filter or no filter (#3073)

2026-09-26 20:44:18 +08:00

router

fix(session): omit meta_info from chat responses (#3877)

2026-10-01 11:45:21 -07:00

scripts

…

tinker

Reach an independently deployed trainer from the launch that drives it (#2936)

2026-09-26 19:29:42 +08:00

tools

Give every worker that watches pods the platform read access it needs (#3056)

2026-09-26 20:36:27 +08:00

true_on_policy

Have every launch script talk to a backend object (#2432)

2026-09-26 18:18:28 +08:00

utils

fix: exec command worker launch commands so actor death frees their ports (#3718)

2026-09-27 20:13:25 -07:00

__init__.py

…

cluster_backends.py

Teach the e2e harness to install a run as several deployments (#3005)

2026-09-26 20:10:43 +08:00

dist_utils.py

…

import_isolation_utils.py

…

source_scan.py

Fence ray to the launcher and check both wires agree (#2620)

2026-09-26 18:55:25 +08:00

test_backend_dimension.py

Give execute_train the launch it is running (#3028)

2026-09-26 20:21:41 +08:00

test_cluster_backends.py

Teach the ft e2e harness which cluster backend it runs on (#2906)

2026-09-26 19:14:22 +08:00

test_docker_image_inputs.py

…

test_dsv4_thd.py

perf(dsv4): balance the CSA indexer across contiguous CP ranks (#3689)

2026-09-29 09:05:40 -07:00

test_glm5_indexer_query_norm.py

fix: normalize GLM-5 indexer queries with shared RMSNorm (#3715)

2026-09-25 14:45:02 -07:00

test_kpool_indexer_packed.py

GLM 5.3 Flash RL Support (#2786)

2026-09-24 20:05:11 -07:00

test_megatron_cli_flags.py

…

test_nvfp4_scope.py

Limit NVFP4 conversion to main language decoder experts (#3638)

2026-09-24 14:56:05 -07:00

test_platform_contract.py

Drop the empty-batch timeout only multi-LoRA v1 ever raised (#3723)

2026-09-26 20:50:59 +08:00

test_qsa_block_sparse_attn.py

Qwen3.8-Flash-Next RL training support (#2777)

2026-09-24 20:02:43 -07:00

test_qsa_indexer_packed.py

Qwen3.8-Flash-Next RL training support (#2777)

2026-09-24 20:02:43 -07:00

test_qsa_indexer_rope.py

Qwen3.8-Flash-Next RL training support (#2777)

2026-09-24 20:02:43 -07:00

test_qsa_tile_index.py

Qwen3.8-Flash-Next RL training support (#2777)

2026-09-24 20:02:43 -07:00

test_qwen3_vl_cp_mrope.py

…

test_qwen_gdn_backend.py

…

test_ray_cluster_bootstrap.py

…

test_ray_launcher_unbuffering.py

…

test_timer_event_sinks.py

…

test_train_async.py

Shut the worker manager down on every driver exit path (#3120)

2026-09-26 20:48:28 +08:00

test_train_multi_policy.py

Shut the worker manager down on every driver exit path (#3120)

2026-09-26 20:48:28 +08:00

test_train.py

Shut the worker manager down on every driver exit path (#3120)

2026-09-26 20:48:28 +08:00

test_training_images.py

Install kubectl and helm into the training image (#2191)

2026-09-26 17:42:02 +08:00

train_args.py

Show a run installed by hand as one deployment per component (#3004)

2026-09-26 20:10:13 +08:00