xpk 0.16.0__tar.gz → 0.16.1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (287) hide show
  1. {xpk-0.16.0/src/xpk.egg-info → xpk-0.16.1}/PKG-INFO +1 -1
  2. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_create.txt +3 -2
  3. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_create_sub-slicing.txt +3 -2
  4. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_create_with_output-manifest-file.txt +3 -2
  5. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/workload.py +1 -0
  6. {xpk-0.16.0 → xpk-0.16.1/src/xpk.egg-info}/PKG-INFO +1 -1
  7. {xpk-0.16.0 → xpk-0.16.1}/.dockerignore +0 -0
  8. {xpk-0.16.0 → xpk-0.16.1}/.github/CODEOWNERS +0 -0
  9. {xpk-0.16.0 → xpk-0.16.1}/.github/PULL_REQUEST_TEMPLATE.md +0 -0
  10. {xpk-0.16.0 → xpk-0.16.1}/.github/actions/install-kjob/action.yml +0 -0
  11. {xpk-0.16.0 → xpk-0.16.1}/.github/actions/install-kueue/action.yml +0 -0
  12. {xpk-0.16.0 → xpk-0.16.1}/.github/actions/setup-test-env/action.yml +0 -0
  13. {xpk-0.16.0 → xpk-0.16.1}/.github/release.yaml +0 -0
  14. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/README.md +0 -0
  15. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/build_tests.yaml +0 -0
  16. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/build_wheels.yaml +0 -0
  17. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/cleanup.yaml +0 -0
  18. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/gemini-dispatch.yml +0 -0
  19. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/gemini-invoke.yml +0 -0
  20. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/gemini-review.yml +0 -0
  21. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/gemini-scheduled-triage.yml +0 -0
  22. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/gemini-triage.yml +0 -0
  23. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/integration_basic_cluster_create.yaml +0 -0
  24. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/integration_legacy_tests.yaml +0 -0
  25. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/integration_pathways_cluster_create.yaml +0 -0
  26. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/integration_ray_cluster_create.yaml +0 -0
  27. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/integration_storage_tests.yaml +0 -0
  28. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/label-validation.yaml +0 -0
  29. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/nightly_tests.yaml +0 -0
  30. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_build_kjob.yaml +0 -0
  31. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_build_scripts.yaml +0 -0
  32. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_build_wheel.yaml +0 -0
  33. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_goldens.yaml +0 -0
  34. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_integration_tests.yaml +0 -0
  35. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_lint_and_format.yml +0 -0
  36. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_storage_create.yaml +0 -0
  37. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_storage_delete.yaml +0 -0
  38. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/reusable_unit_tests.yaml +0 -0
  39. {xpk-0.16.0 → xpk-0.16.1}/.github/workflows/stale.yaml +0 -0
  40. {xpk-0.16.0 → xpk-0.16.1}/.gitignore +0 -0
  41. {xpk-0.16.0 → xpk-0.16.1}/.pre-commit-config.yaml +0 -0
  42. {xpk-0.16.0 → xpk-0.16.1}/LICENSE +0 -0
  43. {xpk-0.16.0 → xpk-0.16.1}/Makefile +0 -0
  44. {xpk-0.16.0 → xpk-0.16.1}/README.md +0 -0
  45. {xpk-0.16.0 → xpk-0.16.1}/backoff_retry.sh +0 -0
  46. {xpk-0.16.0 → xpk-0.16.1}/data/Dockerfile +0 -0
  47. {xpk-0.16.0 → xpk-0.16.1}/docs/code-of-conduct.md +0 -0
  48. {xpk-0.16.0 → xpk-0.16.1}/docs/contributing.md +0 -0
  49. {xpk-0.16.0 → xpk-0.16.1}/docs/installation.md +0 -0
  50. {xpk-0.16.0 → xpk-0.16.1}/docs/local_testing.md +0 -0
  51. {xpk-0.16.0 → xpk-0.16.1}/docs/permissions.md +0 -0
  52. {xpk-0.16.0 → xpk-0.16.1}/docs/testing.md +0 -0
  53. {xpk-0.16.0 → xpk-0.16.1}/docs/troubleshooting.md +0 -0
  54. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/advanced.md +0 -0
  55. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/autoprovisioning.md +0 -0
  56. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/clusters.md +0 -0
  57. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/cpu.md +0 -0
  58. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/docker.md +0 -0
  59. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/gpu.md +0 -0
  60. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/inspector.md +0 -0
  61. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/job.md +0 -0
  62. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/run.md +0 -0
  63. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/storage.md +0 -0
  64. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/tpu7x/clusters.md +0 -0
  65. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/tpu7x/recipes/flex_filestore_recipe.md +0 -0
  66. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/tpu7x/recipes/flex_lustre_recipe.md +0 -0
  67. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/tpu7x/recipes/reservation_gcs_bucket_recipe.md +0 -0
  68. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/tpu7x/workloads.md +0 -0
  69. {xpk-0.16.0 → xpk-0.16.1}/docs/usage/workloads.md +0 -0
  70. {xpk-0.16.0 → xpk-0.16.1}/examples/batch.md +0 -0
  71. {xpk-0.16.0 → xpk-0.16.1}/examples/fake_training.py +0 -0
  72. {xpk-0.16.0 → xpk-0.16.1}/examples/job.sh +0 -0
  73. {xpk-0.16.0 → xpk-0.16.1}/examples/llama-3.1-finetuning/check_cuda.sh +0 -0
  74. {xpk-0.16.0 → xpk-0.16.1}/examples/llama-3.1-finetuning/requirements.txt +0 -0
  75. {xpk-0.16.0 → xpk-0.16.1}/examples/llama-3.1-finetuning/train.py +0 -0
  76. {xpk-0.16.0 → xpk-0.16.1}/examples/llama-3.1-finetuning/train.slurm +0 -0
  77. {xpk-0.16.0 → xpk-0.16.1}/examples/llama-3.1-finetuning/training_data.jsonl +0 -0
  78. {xpk-0.16.0 → xpk-0.16.1}/examples/nccl/nccl-a3mega.sh +0 -0
  79. {xpk-0.16.0 → xpk-0.16.1}/examples/nccl/nccl-a3ultra.sh +0 -0
  80. {xpk-0.16.0 → xpk-0.16.1}/examples/nccl/nccl.md +0 -0
  81. {xpk-0.16.0 → xpk-0.16.1}/examples/storage/filestore-manifest-attach.yaml +0 -0
  82. {xpk-0.16.0 → xpk-0.16.1}/examples/storage/gcsfuse-manifest.yaml +0 -0
  83. {xpk-0.16.0 → xpk-0.16.1}/examples/storage/lustre-manifest-attach.yaml +0 -0
  84. {xpk-0.16.0 → xpk-0.16.1}/examples/storage/parallelstore-manifest-attach.yaml +0 -0
  85. {xpk-0.16.0 → xpk-0.16.1}/examples/storage/pd-manifest-attach.yaml +0 -0
  86. {xpk-0.16.0 → xpk-0.16.1}/golden_buddy.sh +0 -0
  87. {xpk-0.16.0 → xpk-0.16.1}/goldens/Basic_cluster_create.txt +0 -0
  88. {xpk-0.16.0 → xpk-0.16.1}/goldens/Batch.txt +0 -0
  89. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_for_multi-host_nodepool.txt +0 -0
  90. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_for_single-host_single-slice_TPU.txt +0 -0
  91. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_private.txt +0 -0
  92. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_sub-slicing.txt +0 -0
  93. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_CPU_and_memory_limits_above_capacity.txt +0 -0
  94. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_CPU_and_memory_limits_below_capacity.txt +0 -0
  95. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_Managed_Lustre_driver.txt +0 -0
  96. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_Managed_Lustre_driver_and_legacy_port.txt +0 -0
  97. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_gb200-4.txt +0 -0
  98. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_create_with_shared_reservation.txt +0 -0
  99. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_delete.txt +0 -0
  100. {xpk-0.16.0 → xpk-0.16.1}/goldens/Cluster_delete_force.txt +0 -0
  101. {xpk-0.16.0 → xpk-0.16.1}/goldens/Job_cancel.txt +0 -0
  102. {xpk-0.16.0 → xpk-0.16.1}/goldens/Job_info.txt +0 -0
  103. {xpk-0.16.0 → xpk-0.16.1}/goldens/Job_list.txt +0 -0
  104. {xpk-0.16.0 → xpk-0.16.1}/goldens/NAP_cluster-create.txt +0 -0
  105. {xpk-0.16.0 → xpk-0.16.1}/goldens/NAP_cluster-create_with_pathways.txt +0 -0
  106. {xpk-0.16.0 → xpk-0.16.1}/goldens/Storage_list.txt +0 -0
  107. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_create_pathways.txt +0 -0
  108. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_delete.txt +0 -0
  109. {xpk-0.16.0 → xpk-0.16.1}/goldens/Workload_list.txt +0 -0
  110. {xpk-0.16.0 → xpk-0.16.1}/goldens.yaml +0 -0
  111. {xpk-0.16.0 → xpk-0.16.1}/pylintrc +0 -0
  112. {xpk-0.16.0 → xpk-0.16.1}/pyproject.toml +0 -0
  113. {xpk-0.16.0 → xpk-0.16.1}/setup.cfg +0 -0
  114. {xpk-0.16.0 → xpk-0.16.1}/src/integration/README.md +0 -0
  115. {xpk-0.16.0 → xpk-0.16.1}/src/integration/__init__.py +0 -0
  116. {xpk-0.16.0 → xpk-0.16.1}/src/integration/docker_manager_test.py +0 -0
  117. {xpk-0.16.0 → xpk-0.16.1}/src/integration/gcluster_a3mega_test.py +0 -0
  118. {xpk-0.16.0 → xpk-0.16.1}/src/integration/gcluster_a3ultra_test.py +0 -0
  119. {xpk-0.16.0 → xpk-0.16.1}/src/integration/gcluster_a4_test.py +0 -0
  120. {xpk-0.16.0 → xpk-0.16.1}/src/integration/gcluster_test.py +0 -0
  121. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/__init__.py +0 -0
  122. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/api/__init__.py +0 -0
  123. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/api/storage_crd.yaml +0 -0
  124. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3mega/config-map.yaml.tftpl +0 -0
  125. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3mega/storage_crd.yaml +0 -0
  126. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3ultra/config-map.yaml.tftpl +0 -0
  127. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3ultra/mlgru-disable.yaml +0 -0
  128. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3ultra/nccl-installer.yaml +0 -0
  129. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a3ultra/storage_crd.yaml +0 -0
  130. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a4/config-map.yaml.tftpl +0 -0
  131. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a4/nccl-rdma-installer-a4.yaml +0 -0
  132. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/blueprints/a4/storage_crd.yaml +0 -0
  133. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/__init__.py +0 -0
  134. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/batch.py +0 -0
  135. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/cluster.py +0 -0
  136. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/cluster_gcluster.py +0 -0
  137. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/cluster_gcluster_test.py +0 -0
  138. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/cluster_test.py +0 -0
  139. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/common.py +0 -0
  140. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/config.py +0 -0
  141. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/info.py +0 -0
  142. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/inspector.py +0 -0
  143. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/job.py +0 -0
  144. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/kind.py +0 -0
  145. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/kjob_common.py +0 -0
  146. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/managed_ml_diagnostics.py +0 -0
  147. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/managed_ml_diagnostics_test.py +0 -0
  148. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/run.py +0 -0
  149. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/shell.py +0 -0
  150. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/storage.py +0 -0
  151. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/version.py +0 -0
  152. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/commands/workload_test.py +0 -0
  153. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/__init__.py +0 -0
  154. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/__init__.py +0 -0
  155. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/blueprint_definitions.py +0 -0
  156. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/blueprint_generator.py +0 -0
  157. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/blueprint_test.py +0 -0
  158. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/testing/__init__.py +0 -0
  159. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/testing/data/a3_mega.yaml +0 -0
  160. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/testing/data/a3_mega_spot.yaml +0 -0
  161. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/testing/data/a3_ultra.yaml +0 -0
  162. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/blueprint/testing/data/a4.yaml +0 -0
  163. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/capacity.py +0 -0
  164. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/capacity_test.py +0 -0
  165. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/cluster.py +0 -0
  166. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/cluster_private.py +0 -0
  167. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/cluster_test.py +0 -0
  168. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/commands.py +0 -0
  169. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/config.py +0 -0
  170. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/config_test.py +0 -0
  171. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/docker_container.py +0 -0
  172. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/docker_image.py +0 -0
  173. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/docker_manager.py +0 -0
  174. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/docker_resources.py +0 -0
  175. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/filestore.py +0 -0
  176. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/gcloud_context.py +0 -0
  177. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/gcloud_context_test.py +0 -0
  178. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/gcluster_manager.py +0 -0
  179. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/gcsfuse.py +0 -0
  180. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/jobset.py +0 -0
  181. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/kjob.py +0 -0
  182. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/kueue_manager.py +0 -0
  183. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/kueue_manager_test.py +0 -0
  184. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/monitoring.py +0 -0
  185. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/mtc.py +0 -0
  186. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/nap.py +0 -0
  187. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/network.py +0 -0
  188. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/nodepool.py +0 -0
  189. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/nodepool_test.py +0 -0
  190. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/pathways.py +0 -0
  191. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/pathways_test.py +0 -0
  192. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/ray.py +0 -0
  193. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/remote_state/__init__.py +0 -0
  194. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/remote_state/fuse_remote_state.py +0 -0
  195. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/remote_state/remote_state_client.py +0 -0
  196. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/resources.py +0 -0
  197. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/scheduling.py +0 -0
  198. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/scheduling_test.py +0 -0
  199. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/storage.py +0 -0
  200. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/system_characteristics.py +0 -0
  201. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/system_characteristics_test.py +0 -0
  202. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/telemetry.py +0 -0
  203. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/telemetry_test.py +0 -0
  204. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/testing/__init__.py +0 -0
  205. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/testing/commands_tester.py +0 -0
  206. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/testing/commands_tester_test.py +0 -0
  207. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/updates.py +0 -0
  208. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/updates_test.py +0 -0
  209. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/vertex.py +0 -0
  210. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload.py +0 -0
  211. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/__init__.py +0 -0
  212. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/rdma_decorator.py +0 -0
  213. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/storage_decorator.py +0 -0
  214. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/tcpx_decorator.py +0 -0
  215. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/tcpx_decorator_test.py +0 -0
  216. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_decorators/tcpxo_decorator.py +0 -0
  217. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/core/workload_test.py +0 -0
  218. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/main.py +0 -0
  219. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/__init__.py +0 -0
  220. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/batch.py +0 -0
  221. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/cluster.py +0 -0
  222. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/cluster_test.py +0 -0
  223. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/common.py +0 -0
  224. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/config.py +0 -0
  225. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/core.py +0 -0
  226. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/info.py +0 -0
  227. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/inspector.py +0 -0
  228. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/job.py +0 -0
  229. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/kind.py +0 -0
  230. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/run.py +0 -0
  231. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/shell.py +0 -0
  232. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/storage.py +0 -0
  233. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/storage_test.py +0 -0
  234. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/validators.py +0 -0
  235. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/version.py +0 -0
  236. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/workload.py +0 -0
  237. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/parser/workload_test.py +0 -0
  238. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/telemetry_uploader.py +0 -0
  239. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/__init__.py +0 -0
  240. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/arm_gpu_workload_crate.yaml.j2 +0 -0
  241. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/cluster_preheat.yaml.j2 +0 -0
  242. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/filestore-pv.yaml +0 -0
  243. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/filestore-pvc.yaml +0 -0
  244. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/filestore-sc.yaml +0 -0
  245. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/fuse-pv.yaml +0 -0
  246. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/fuse-pvc.yaml +0 -0
  247. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/kueue_config.yaml.j2 +0 -0
  248. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/kueue_gke_default_topology.yaml.j2 +0 -0
  249. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/kueue_sub_slicing_topology.yaml.j2 +0 -0
  250. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/mtc-cpc.yaml +0 -0
  251. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/storage.yaml +0 -0
  252. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/templates/volume_bundle.yaml +0 -0
  253. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/__init__.py +0 -0
  254. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/console.py +0 -0
  255. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/console_test.py +0 -0
  256. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/execution_context.py +0 -0
  257. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/feature_flags.py +0 -0
  258. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/file.py +0 -0
  259. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/gcs_utils.py +0 -0
  260. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/kubectl.py +0 -0
  261. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/kueue.py +0 -0
  262. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/network.py +0 -0
  263. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/objects.py +0 -0
  264. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/templates.py +0 -0
  265. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/topology.py +0 -0
  266. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/topology_test.py +0 -0
  267. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/user_agent.py +0 -0
  268. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/user_agent_test.py +0 -0
  269. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/user_input.py +0 -0
  270. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/user_input_test.py +0 -0
  271. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/validation.py +0 -0
  272. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/validation_test.py +0 -0
  273. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/versions.py +0 -0
  274. {xpk-0.16.0 → xpk-0.16.1}/src/xpk/utils/yaml.py +0 -0
  275. {xpk-0.16.0 → xpk-0.16.1}/src/xpk.egg-info/SOURCES.txt +0 -0
  276. {xpk-0.16.0 → xpk-0.16.1}/src/xpk.egg-info/dependency_links.txt +0 -0
  277. {xpk-0.16.0 → xpk-0.16.1}/src/xpk.egg-info/entry_points.txt +0 -0
  278. {xpk-0.16.0 → xpk-0.16.1}/src/xpk.egg-info/requires.txt +0 -0
  279. {xpk-0.16.0 → xpk-0.16.1}/src/xpk.egg-info/top_level.txt +0 -0
  280. {xpk-0.16.0 → xpk-0.16.1}/tools/Dockerfile-kjob +0 -0
  281. {xpk-0.16.0 → xpk-0.16.1}/tools/build-kjob.sh +0 -0
  282. {xpk-0.16.0 → xpk-0.16.1}/tools/install-gke-auth-plugin.sh +0 -0
  283. {xpk-0.16.0 → xpk-0.16.1}/tools/install-xpk.sh +0 -0
  284. {xpk-0.16.0 → xpk-0.16.1}/xpk-large-scale-guide.sh +0 -0
  285. {xpk-0.16.0 → xpk-0.16.1}/xpk-notebooks.md +0 -0
  286. {xpk-0.16.0 → xpk-0.16.1}/xpk-slurm-commands.md +0 -0
  287. {xpk-0.16.0 → xpk-0.16.1}/xpk.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: xpk
3
- Version: 0.16.0
3
+ Version: 0.16.1
4
4
  Summary: xpk helps Cloud developers to orchestrate training jobs on accelerators on GKE.
5
5
  Author-email: XPK team <xpk-code-reviewers@google.com>
6
6
  License: Apache-2.0
@@ -35,7 +35,7 @@ docker buildx build --platform=linux/amd64 -f 4b6736a12db8ea0f78ce793fd0d4ee0c94
35
35
  docker tag dry-run-runner gcr.io/golden-project/dry-run-runner:prefix-current
36
36
  [XPK] Task: `Upload Docker Image` is implemented by the following command not running since it is a dry run.
37
37
  docker push gcr.io/golden-project/dry-run-runner:prefix-current
38
- [XPK] Temp file (e21c8ebdc21d15a852187058c096898c486d3b1066e67dcfb67e5052a1d0a7fa) content:
38
+ [XPK] Temp file (d839cb02cf17813401b99c09dcf1cf289c44511f39e964d50d4b3e8d11eaa2da) content:
39
39
  apiVersion: jobset.x-k8s.io/v1alpha2
40
40
  kind: JobSet
41
41
  metadata:
@@ -65,6 +65,7 @@ spec:
65
65
  podFailurePolicy:
66
66
  rules:
67
67
  - action: FailJob
68
+ onPodConditions: []
68
69
  onExitCodes:
69
70
  containerName: jax-tpu
70
71
  operator: NotIn
@@ -144,7 +145,7 @@ spec:
144
145
 
145
146
 
146
147
  [XPK] Task: `Creating Workload` is implemented by the following command not running since it is a dry run.
147
- kubectl apply -f e21c8ebdc21d15a852187058c096898c486d3b1066e67dcfb67e5052a1d0a7fa
148
+ kubectl apply -f d839cb02cf17813401b99c09dcf1cf289c44511f39e964d50d4b3e8d11eaa2da
148
149
  [XPK] Task: `GKE Dashboard List` is implemented by the following command not running since it is a dry run.
149
150
  gcloud monitoring dashboards list --project=golden-project --filter="displayName:'GKE - TPU Monitoring Dashboard'" --format="value(name)" --verbosity=error
150
151
  [XPK] Check statistics and outlier mode of GKE metrics here: https://console.cloud.google.com/monitoring/dashboards/builder/0?project=golden-project&f.rlabel.cluster_name.ClusterName=golden-cluster. To view the metric data for your workload, select golden-workload from the JobName filter on the dashboard.
@@ -39,7 +39,7 @@ docker buildx build --platform=linux/amd64 -f 4b6736a12db8ea0f78ce793fd0d4ee0c94
39
39
  docker tag dry-run-runner gcr.io/golden-project/dry-run-runner:prefix-current
40
40
  [XPK] Task: `Upload Docker Image` is implemented by the following command not running since it is a dry run.
41
41
  docker push gcr.io/golden-project/dry-run-runner:prefix-current
42
- [XPK] Temp file (1fd31adc18f9b13689304d4254726dbbab4ac259cbdfe368bb4edd2bbd270f16) content:
42
+ [XPK] Temp file (d93ff5dcd4d42cc22ebbdc043d7dc82f00c01be2222751ef9fba03fc8079ba68) content:
43
43
  apiVersion: jobset.x-k8s.io/v1alpha2
44
44
  kind: JobSet
45
45
  metadata:
@@ -69,6 +69,7 @@ spec:
69
69
  podFailurePolicy:
70
70
  rules:
71
71
  - action: FailJob
72
+ onPodConditions: []
72
73
  onExitCodes:
73
74
  containerName: jax-tpu
74
75
  operator: NotIn
@@ -149,7 +150,7 @@ spec:
149
150
 
150
151
 
151
152
  [XPK] Task: `Creating Workload` is implemented by the following command not running since it is a dry run.
152
- kubectl apply -f 1fd31adc18f9b13689304d4254726dbbab4ac259cbdfe368bb4edd2bbd270f16
153
+ kubectl apply -f d93ff5dcd4d42cc22ebbdc043d7dc82f00c01be2222751ef9fba03fc8079ba68
153
154
  [XPK] Task: `GKE Dashboard List` is implemented by the following command not running since it is a dry run.
154
155
  gcloud monitoring dashboards list --project=golden-project --filter="displayName:'GKE - TPU Monitoring Dashboard'" --format="value(name)" --verbosity=error
155
156
  [XPK] Check statistics and outlier mode of GKE metrics here: https://console.cloud.google.com/monitoring/dashboards/builder/0?project=golden-project&f.rlabel.cluster_name.ClusterName=golden-cluster. To view the metric data for your workload, select golden-workload from the JobName filter on the dashboard.
@@ -36,7 +36,7 @@ docker tag dry-run-runner gcr.io/golden-project/dry-run-runner:prefix-current
36
36
  [XPK] Task: `Upload Docker Image` is implemented by the following command not running since it is a dry run.
37
37
  docker push gcr.io/golden-project/dry-run-runner:prefix-current
38
38
  [XPK] Workload golden-workload manifest written to /var/tmp/manifest.yaml
39
- [XPK] Temp file (e21c8ebdc21d15a852187058c096898c486d3b1066e67dcfb67e5052a1d0a7fa) content:
39
+ [XPK] Temp file (d839cb02cf17813401b99c09dcf1cf289c44511f39e964d50d4b3e8d11eaa2da) content:
40
40
  apiVersion: jobset.x-k8s.io/v1alpha2
41
41
  kind: JobSet
42
42
  metadata:
@@ -66,6 +66,7 @@ spec:
66
66
  podFailurePolicy:
67
67
  rules:
68
68
  - action: FailJob
69
+ onPodConditions: []
69
70
  onExitCodes:
70
71
  containerName: jax-tpu
71
72
  operator: NotIn
@@ -145,7 +146,7 @@ spec:
145
146
 
146
147
 
147
148
  [XPK] Task: `Creating Workload` is implemented by the following command not running since it is a dry run.
148
- kubectl apply -f e21c8ebdc21d15a852187058c096898c486d3b1066e67dcfb67e5052a1d0a7fa
149
+ kubectl apply -f d839cb02cf17813401b99c09dcf1cf289c44511f39e964d50d4b3e8d11eaa2da
149
150
  [XPK] Task: `GKE Dashboard List` is implemented by the following command not running since it is a dry run.
150
151
  gcloud monitoring dashboards list --project=golden-project --filter="displayName:'GKE - TPU Monitoring Dashboard'" --format="value(name)" --verbosity=error
151
152
  [XPK] Check statistics and outlier mode of GKE metrics here: https://console.cloud.google.com/monitoring/dashboards/builder/0?project=golden-project&f.rlabel.cluster_name.ClusterName=golden-cluster. To view the metric data for your workload, select golden-workload from the JobName filter on the dashboard.
@@ -492,6 +492,7 @@ def workload_create(args) -> None:
492
492
  podFailurePolicy:
493
493
  rules:
494
494
  - action: FailJob
495
+ onPodConditions: []
495
496
  onExitCodes:
496
497
  containerName: {get_main_container_docker_image(args, workload_system)}
497
498
  operator: NotIn
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: xpk
3
- Version: 0.16.0
3
+ Version: 0.16.1
4
4
  Summary: xpk helps Cloud developers to orchestrate training jobs on accelerators on GKE.
5
5
  Author-email: XPK team <xpk-code-reviewers@google.com>
6
6
  License: Apache-2.0
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes
File without changes