|
15 | 15 | """DAGs to run Aotc reproducibility benchmarks."""
|
16 | 16 |
|
17 | 17 | import datetime
|
| 18 | +import os |
18 | 19 |
|
19 | 20 | from airflow import models
|
20 |
| -from dags.map_reproducibility.utils.constants import Schedule |
| 21 | +from dags.map_reproducibility.utils.constants import Schedule, Image |
21 | 22 | from dags.map_reproducibility.utils.internal_aotc_workload import run_internal_aotc_workload
|
22 | 23 |
|
23 | 24 |
|
| 25 | +# Configuration parameters |
24 | 26 | TEST_RUN = False
|
25 | 27 | TURN_ON_SCHEDULE = True
|
26 | 28 | BACKFILL = False
|
27 | 29 |
|
28 |
| -# List of configuration setups as a dictionary with schedule times |
29 |
| -config_yamls = { |
30 |
| - # a3ultra_llama3.1-8b |
31 |
| - "recipes/a3ultra/a3ultra_llama3.1-8b_8gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, # < 10mins |
32 |
| - "recipes/a3ultra/a3ultra_llama3.1-8b_8gpus_fp8_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, |
33 |
| - "recipes/a3ultra/a3ultra_llama3.1-8b_16gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, |
34 |
| - "recipes/a3ultra/a3ultra_llama3.1-8b_16gpus_fp8_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, |
35 |
| - # a3ultra_mixtral-8x7 |
36 |
| - "recipes/a3ultra/a3ultra_mixtral-8x7b_8gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, |
37 |
| - "recipes/a3ultra/a3ultra_mixtral-8x7b_16gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_6PM_EXCEPT_THURSDAY, |
38 |
| - # a3ultra_llama3.1-70b |
39 |
| - "recipes/a3ultra/a3ultra_llama3.1-70b_256gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_6_30PM_EXCEPT_THURSDAY, # ~10min |
40 |
| - "recipes/a3ultra/a3ultra_llama3.1-70b_256gpus_fp8_maxtext.yaml": Schedule.DAILY_PST_6_30PM_EXCEPT_THURSDAY, |
41 |
| - # a3ultra_llama3.1-405b |
42 |
| - "recipes/a3ultra/a3ultra_llama3.1-405b_256gpus_fp8_maxtext.yaml": Schedule.DAILY_PST_7PM_EXCEPT_THURSDAY, # ~30mins |
43 |
| - "recipes/a3ultra/a3ultra_llama3.1-405b_256gpus_bf16_maxtext.yaml": Schedule.DAILY_PST_7_30PM_EXCEPT_THURSDAY, # ~30mins |
44 |
| - # Add more config paths as needed |
45 |
| -} |
| 30 | +# Get current date for image tags |
| 31 | +utc_date = datetime.datetime.now(datetime.timezone.utc).strftime("%Y-%m-%d") |
| 32 | +NIGHTLY_IMAGE = f"{Image.MAXTEXT_JAX_STABLE_NIGHTLY}:{utc_date}" |
| 33 | +RELEASE_IMAGE = f"{Image.MAXTEXT_JAX_STABLE_RELEASE}:{utc_date}" |
46 | 34 |
|
47 |
| -# Define common tags |
48 |
| -common_tags = [ |
| 35 | +# Common DAG tags |
| 36 | +DAG_TAGS = [ |
49 | 37 | "reproducibility",
|
50 | 38 | "experimental",
|
51 | 39 | "xlml",
|
52 |
| - "v1.15", |
| 40 | + "v1.16", |
53 | 41 | "internal",
|
54 | 42 | "regressiontests",
|
55 | 43 | "a3ultra",
|
56 | 44 | ]
|
57 | 45 |
|
58 |
| -# Create a DAG for each config |
59 |
| -for relative_config_yaml_path, schedule_time in config_yamls.items(): |
| 46 | +# Model configurations with schedule and timeout settings |
| 47 | +MODEL_CONFIGS = { |
| 48 | + # a3ultra_llama3.1-8b |
| 49 | + "recipes/a3ultra/a3ultra_llama3.1-8b_8gpus_bf16_maxtext.yaml": { |
| 50 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 51 | + "timeout_minutes": 15, |
| 52 | + }, |
| 53 | + "recipes/a3ultra/a3ultra_llama3.1-8b_8gpus_fp8_maxtext.yaml": { |
| 54 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 55 | + "timeout_minutes": 15, |
| 56 | + }, |
| 57 | + "recipes/a3ultra/a3ultra_llama3.1-8b_16gpus_bf16_maxtext.yaml": { |
| 58 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 59 | + "timeout_minutes": 15, |
| 60 | + }, |
| 61 | + "recipes/a3ultra/a3ultra_llama3.1-8b_16gpus_fp8_maxtext.yaml": { |
| 62 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 63 | + "timeout_minutes": 15, |
| 64 | + }, |
| 65 | + # a3ultra_mixtral-8x7 |
| 66 | + "recipes/a3ultra/a3ultra_mixtral-8x7b_8gpus_bf16_maxtext.yaml": { |
| 67 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 68 | + "timeout_minutes": 15, |
| 69 | + }, |
| 70 | + "recipes/a3ultra/a3ultra_mixtral-8x7b_16gpus_bf16_maxtext.yaml": { |
| 71 | + "schedule": Schedule.DAILY_PDT_6PM_EXCEPT_THURSDAY, |
| 72 | + "timeout_minutes": 15, |
| 73 | + }, |
| 74 | + # a3ultra_llama3.1-70b |
| 75 | + "recipes/a3ultra/a3ultra_llama3.1-70b_256gpus_bf16_maxtext.yaml": { |
| 76 | + "schedule": Schedule.DAILY_PDT_6_30PM_EXCEPT_THURSDAY, |
| 77 | + "timeout_minutes": 15, |
| 78 | + }, |
| 79 | + "recipes/a3ultra/a3ultra_llama3.1-70b_256gpus_fp8_maxtext.yaml": { |
| 80 | + "schedule": Schedule.DAILY_PDT_7PM_EXCEPT_THURSDAY, |
| 81 | + "timeout_minutes": 15, |
| 82 | + }, |
| 83 | + # a3ultra_llama3.1-405b |
| 84 | + "recipes/a3ultra/a3ultra_llama3.1-405b_256gpus_fp8_maxtext.yaml": { |
| 85 | + "schedule": Schedule.DAILY_PDT_7_30PM_EXCEPT_THURSDAY, |
| 86 | + "timeout_minutes": 30, |
| 87 | + }, |
| 88 | + "recipes/a3ultra/a3ultra_llama3.1-405b_256gpus_bf16_maxtext.yaml": { |
| 89 | + "schedule": Schedule.DAILY_PDT_8PM_EXCEPT_THURSDAY, |
| 90 | + "timeout_minutes": 40, |
| 91 | + }, |
| 92 | +} |
| 93 | + |
| 94 | + |
| 95 | +# Create DAGs for each configuration |
| 96 | +for config_path, config_info in MODEL_CONFIGS.items(): |
60 | 97 | # Extract config name for the DAG ID
|
61 |
| - config_yaml_name = relative_config_yaml_path.rsplit("/", maxsplit=1)[ |
62 |
| - -1 |
63 |
| - ].replace(".yaml", "") |
64 |
| - actual_schedule = schedule_time if TURN_ON_SCHEDULE else None |
65 |
| - dag_id = f"new_internal_{config_yaml_name}" |
| 98 | + config_name = os.path.basename(config_path).replace(".yaml", "") |
| 99 | + schedule = config_info["schedule"] if TURN_ON_SCHEDULE else None |
| 100 | + timeout = config_info["timeout_minutes"] |
| 101 | + |
| 102 | + # Create DAG for nightly build |
| 103 | + with models.DAG( |
| 104 | + dag_id=f"new_internal_{config_name}", |
| 105 | + schedule=schedule, |
| 106 | + tags=DAG_TAGS, |
| 107 | + start_date=datetime.datetime(2025, 4, 3), |
| 108 | + catchup=False, |
| 109 | + ) as dag: |
| 110 | + run_internal_aotc_workload( |
| 111 | + relative_config_yaml_path=config_path, |
| 112 | + test_run=TEST_RUN, |
| 113 | + backfill=BACKFILL, |
| 114 | + timeout=timeout, |
| 115 | + image_version=NIGHTLY_IMAGE, |
| 116 | + ) |
66 | 117 |
|
67 |
| - # Define the DAG |
| 118 | + # Create DAG for stable release |
68 | 119 | with models.DAG(
|
69 |
| - dag_id=dag_id, |
70 |
| - schedule=actual_schedule, # Use the specific schedule time |
71 |
| - tags=common_tags, |
| 120 | + dag_id=f"new_internal_stable_release_{config_name}_", |
| 121 | + schedule=schedule, |
| 122 | + tags=DAG_TAGS, |
72 | 123 | start_date=datetime.datetime(2025, 4, 3),
|
73 | 124 | catchup=False,
|
74 | 125 | ) as dag:
|
75 |
| - # Create the workload for this specific config |
76 | 126 | run_internal_aotc_workload(
|
77 |
| - relative_config_yaml_path=relative_config_yaml_path, |
| 127 | + relative_config_yaml_path=config_path, |
78 | 128 | test_run=TEST_RUN,
|
79 | 129 | backfill=BACKFILL,
|
| 130 | + timeout=timeout, |
| 131 | + image_version=RELEASE_IMAGE, |
80 | 132 | )
|
0 commit comments