SYNKA 26805: verified launches/SYNKA-26805/synka-p2-15c46844396f00d2034b7a1e-26805-33c136b1.json
Browse files
launches/SYNKA-26805/synka-p2-15c46844396f00d2034b7a1e-26805-33c136b1.json
ADDED
|
@@ -0,0 +1,265 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_commit": "afb227e73655d0077702d9e5203fa524fe09b56a",
|
| 3 |
+
"original_spec": {
|
| 4 |
+
"environment": "looptrain",
|
| 5 |
+
"module": "src.pretrain.train_titan",
|
| 6 |
+
"arguments": [
|
| 7 |
+
"--arch",
|
| 8 |
+
"P2",
|
| 9 |
+
"--run-name",
|
| 10 |
+
"P2",
|
| 11 |
+
"--data",
|
| 12 |
+
"@TOKENIZED_ROOT@",
|
| 13 |
+
"--seed",
|
| 14 |
+
"42",
|
| 15 |
+
"--verify-sha256",
|
| 16 |
+
"--on-boundary",
|
| 17 |
+
"raise",
|
| 18 |
+
"--trace-every",
|
| 19 |
+
"100",
|
| 20 |
+
"--activation_checkpoint.mode",
|
| 21 |
+
"none",
|
| 22 |
+
"--activation_checkpoint.selective_ac_option",
|
| 23 |
+
"2",
|
| 24 |
+
"--job.dump_folder",
|
| 25 |
+
"@RUN_OUT@",
|
| 26 |
+
"--training.steps",
|
| 27 |
+
"50000",
|
| 28 |
+
"--training.local_batch_size",
|
| 29 |
+
"12",
|
| 30 |
+
"--training.global_batch_size",
|
| 31 |
+
"96",
|
| 32 |
+
"--training.mixed_precision_param",
|
| 33 |
+
"bfloat16",
|
| 34 |
+
"--training.mixed_precision_reduce",
|
| 35 |
+
"float32",
|
| 36 |
+
"--optimizer.name",
|
| 37 |
+
"AdamW",
|
| 38 |
+
"--optimizer.lr",
|
| 39 |
+
"0.0003",
|
| 40 |
+
"--optimizer.beta1",
|
| 41 |
+
"0.9",
|
| 42 |
+
"--optimizer.beta2",
|
| 43 |
+
"0.95",
|
| 44 |
+
"--optimizer.weight_decay",
|
| 45 |
+
"0.1",
|
| 46 |
+
"--training.max_norm",
|
| 47 |
+
"1.0",
|
| 48 |
+
"--lr_scheduler.warmup_steps",
|
| 49 |
+
"1000",
|
| 50 |
+
"--lr_scheduler.decay_type",
|
| 51 |
+
"sqrt",
|
| 52 |
+
"--lr_scheduler.min_lr_factor",
|
| 53 |
+
"0.05",
|
| 54 |
+
"--lr_scheduler.decay_ratio",
|
| 55 |
+
"0.1",
|
| 56 |
+
"--debug.seed",
|
| 57 |
+
"42",
|
| 58 |
+
"--parallelism.data_parallel_shard_degree",
|
| 59 |
+
"8",
|
| 60 |
+
"--checkpoint.enable",
|
| 61 |
+
"--checkpoint.interval",
|
| 62 |
+
"500",
|
| 63 |
+
"--checkpoint.keep_latest_k",
|
| 64 |
+
"11",
|
| 65 |
+
"--checkpoint.no-last_save_model_only",
|
| 66 |
+
"--metrics.enable_wandb",
|
| 67 |
+
"--checkpoint.initial_load_path",
|
| 68 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/coldstarts/P2/482c13b454a93d2d4e63565d98fa16b954d62f5e/coldstart/ABCI-H200-2322705/step-0"
|
| 69 |
+
]
|
| 70 |
+
},
|
| 71 |
+
"resolved_argv": [
|
| 72 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/conda/envs/looptrain/bin/python",
|
| 73 |
+
"-m",
|
| 74 |
+
"torch.distributed.run",
|
| 75 |
+
"--standalone",
|
| 76 |
+
"--nproc_per_node=8",
|
| 77 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/runs/synka-p2-15c46844396f00d2034b7a1e-26805-33c136b1/adapter/python_module.py",
|
| 78 |
+
"src.pretrain.train_titan",
|
| 79 |
+
"--arch",
|
| 80 |
+
"P2",
|
| 81 |
+
"--run-name",
|
| 82 |
+
"P2",
|
| 83 |
+
"--data",
|
| 84 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/looped_moe_local/data/tokenized/sample-100BT.smollm2",
|
| 85 |
+
"--seed",
|
| 86 |
+
"42",
|
| 87 |
+
"--verify-sha256",
|
| 88 |
+
"--on-boundary",
|
| 89 |
+
"raise",
|
| 90 |
+
"--trace-every",
|
| 91 |
+
"100",
|
| 92 |
+
"--activation_checkpoint.mode",
|
| 93 |
+
"none",
|
| 94 |
+
"--activation_checkpoint.selective_ac_option",
|
| 95 |
+
"2",
|
| 96 |
+
"--job.dump_folder",
|
| 97 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/runs/synka-p2-15c46844396f00d2034b7a1e-26805-33c136b1/output",
|
| 98 |
+
"--training.steps",
|
| 99 |
+
"50000",
|
| 100 |
+
"--training.local_batch_size",
|
| 101 |
+
"12",
|
| 102 |
+
"--training.global_batch_size",
|
| 103 |
+
"96",
|
| 104 |
+
"--training.mixed_precision_param",
|
| 105 |
+
"bfloat16",
|
| 106 |
+
"--training.mixed_precision_reduce",
|
| 107 |
+
"float32",
|
| 108 |
+
"--optimizer.name",
|
| 109 |
+
"AdamW",
|
| 110 |
+
"--optimizer.lr",
|
| 111 |
+
"0.0003",
|
| 112 |
+
"--optimizer.beta1",
|
| 113 |
+
"0.9",
|
| 114 |
+
"--optimizer.beta2",
|
| 115 |
+
"0.95",
|
| 116 |
+
"--optimizer.weight_decay",
|
| 117 |
+
"0.1",
|
| 118 |
+
"--training.max_norm",
|
| 119 |
+
"1.0",
|
| 120 |
+
"--lr_scheduler.warmup_steps",
|
| 121 |
+
"1000",
|
| 122 |
+
"--lr_scheduler.decay_type",
|
| 123 |
+
"sqrt",
|
| 124 |
+
"--lr_scheduler.min_lr_factor",
|
| 125 |
+
"0.05",
|
| 126 |
+
"--lr_scheduler.decay_ratio",
|
| 127 |
+
"0.1",
|
| 128 |
+
"--debug.seed",
|
| 129 |
+
"42",
|
| 130 |
+
"--parallelism.data_parallel_shard_degree",
|
| 131 |
+
"8",
|
| 132 |
+
"--checkpoint.enable",
|
| 133 |
+
"--checkpoint.interval",
|
| 134 |
+
"500",
|
| 135 |
+
"--checkpoint.keep_latest_k",
|
| 136 |
+
"11",
|
| 137 |
+
"--checkpoint.no-last_save_model_only",
|
| 138 |
+
"--metrics.enable_wandb",
|
| 139 |
+
"--checkpoint.initial_load_path",
|
| 140 |
+
"/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/coldstarts/P2/482c13b454a93d2d4e63565d98fa16b954d62f5e/coldstart/ABCI-H200-2322705/step-0"
|
| 141 |
+
],
|
| 142 |
+
"environment": {
|
| 143 |
+
"python": "3.11.15 | packaged by conda-forge | (main, Aug 11 2026, 10:26:29) [GCC 14.4.0]",
|
| 144 |
+
"packages": {
|
| 145 |
+
"markdown-it-py": "4.2.0",
|
| 146 |
+
"nvidia-nvtx-cu12": "12.6.77",
|
| 147 |
+
"pluggy": "1.6.0",
|
| 148 |
+
"docstring_parser": "0.18.0",
|
| 149 |
+
"typeguard": "4.6.0",
|
| 150 |
+
"httpx": "0.28.1",
|
| 151 |
+
"tomli": "2.4.1",
|
| 152 |
+
"nvidia-nvshmem-cu12": "3.4.5",
|
| 153 |
+
"nvidia-cusparselt-cu12": "0.7.1",
|
| 154 |
+
"mpmath": "1.3.0",
|
| 155 |
+
"aiosignal": "1.4.0",
|
| 156 |
+
"sympy": "1.14.0",
|
| 157 |
+
"mdurl": "0.1.2",
|
| 158 |
+
"python-dateutil": "2.9.0.post0",
|
| 159 |
+
"grpcio": "1.83.0",
|
| 160 |
+
"nvidia-cufft-cu12": "11.3.0.4",
|
| 161 |
+
"certifi": "2026.7.22",
|
| 162 |
+
"opentelemetry-api": "1.44.0",
|
| 163 |
+
"aiohttp": "3.14.3",
|
| 164 |
+
"setuptools": "78.1.0",
|
| 165 |
+
"requests": "2.34.2",
|
| 166 |
+
"anyio": "4.14.2",
|
| 167 |
+
"nvidia-cuda-nvrtc-cu12": "12.6.85",
|
| 168 |
+
"propcache": "0.5.2",
|
| 169 |
+
"transformers": "5.15.0",
|
| 170 |
+
"tokenizers": "0.22.2",
|
| 171 |
+
"packaging": "26.2",
|
| 172 |
+
"MarkupSafe": "3.0.3",
|
| 173 |
+
"numpy": "2.4.6",
|
| 174 |
+
"h11": "0.16.0",
|
| 175 |
+
"Werkzeug": "3.1.8",
|
| 176 |
+
"fsspec": "2026.2.0",
|
| 177 |
+
"hf-xet": "1.6.0",
|
| 178 |
+
"nvidia-nvjitlink-cu12": "12.6.85",
|
| 179 |
+
"idna": "3.18",
|
| 180 |
+
"nvidia-cusolver-cu12": "11.7.1.2",
|
| 181 |
+
"nvidia-cuda-runtime-cu12": "12.6.77",
|
| 182 |
+
"cuda-pathfinder": "1.2.2",
|
| 183 |
+
"rich": "15.0.0",
|
| 184 |
+
"six": "1.17.0",
|
| 185 |
+
"typing_extensions": "4.15.0",
|
| 186 |
+
"yarl": "1.24.5",
|
| 187 |
+
"platformdirs": "4.11.3",
|
| 188 |
+
"sentry-sdk": "2.68.0",
|
| 189 |
+
"tqdm": "4.70.0",
|
| 190 |
+
"wheel": "0.47.0",
|
| 191 |
+
"annotated-doc": "0.0.5",
|
| 192 |
+
"frozenlist": "1.8.0",
|
| 193 |
+
"torchdata": "0.11.0",
|
| 194 |
+
"protobuf": "7.35.1",
|
| 195 |
+
"nvidia-curand-cu12": "10.3.7.77",
|
| 196 |
+
"PyYAML": "6.0.3",
|
| 197 |
+
"nvidia-cufile-cu12": "1.11.1.6",
|
| 198 |
+
"networkx": "3.6.1",
|
| 199 |
+
"attrs": "26.1.0",
|
| 200 |
+
"tensorboard-data-server": "0.7.2",
|
| 201 |
+
"shellingham": "1.5.4",
|
| 202 |
+
"safetensors": "0.8.0",
|
| 203 |
+
"xxhash": "3.8.1",
|
| 204 |
+
"torch": "2.12.0+cu126",
|
| 205 |
+
"click": "8.4.2",
|
| 206 |
+
"nvidia-cusparse-cu12": "12.5.4.2",
|
| 207 |
+
"charset-normalizer": "3.4.9",
|
| 208 |
+
"absl-py": "2.5.0",
|
| 209 |
+
"datasets": "4.7.0",
|
| 210 |
+
"dill": "0.4.0",
|
| 211 |
+
"torchtitan": "0.2.2",
|
| 212 |
+
"regex": "2026.7.19",
|
| 213 |
+
"tensorboard": "2.21.0",
|
| 214 |
+
"cuda-toolkit": "12.6.3",
|
| 215 |
+
"pytest": "9.1.1",
|
| 216 |
+
"multidict": "6.7.1",
|
| 217 |
+
"pydantic_core": "2.46.4",
|
| 218 |
+
"nvidia-cublas-cu12": "12.6.4.1",
|
| 219 |
+
"einops": "0.8.2",
|
| 220 |
+
"iniconfig": "2.3.0",
|
| 221 |
+
"pip": "26.2.1",
|
| 222 |
+
"huggingface_hub": "1.27.0",
|
| 223 |
+
"Markdown": "3.10.3",
|
| 224 |
+
"annotated-types": "0.8.0",
|
| 225 |
+
"multiprocess": "0.70.18",
|
| 226 |
+
"typer": "0.27.1",
|
| 227 |
+
"nvidia-cudnn-cu12": "9.10.2.21",
|
| 228 |
+
"megatron-core": "0.16.1",
|
| 229 |
+
"urllib3": "2.7.0",
|
| 230 |
+
"aiohappyeyeballs": "2.7.1",
|
| 231 |
+
"cuda-bindings": "12.9.4",
|
| 232 |
+
"Pygments": "2.20.0",
|
| 233 |
+
"nvidia-cuda-cupti-cu12": "12.6.80",
|
| 234 |
+
"tyro": "1.0.15",
|
| 235 |
+
"wandb": "0.28.2",
|
| 236 |
+
"pydantic": "2.13.4",
|
| 237 |
+
"nvidia-nccl-cu12": "2.29.3",
|
| 238 |
+
"pandas": "3.0.5",
|
| 239 |
+
"httpcore": "1.0.9",
|
| 240 |
+
"pyarrow": "25.0.1",
|
| 241 |
+
"Jinja2": "3.1.6",
|
| 242 |
+
"filelock": "3.29.0",
|
| 243 |
+
"typing-inspection": "0.4.4",
|
| 244 |
+
"pillow": "12.3.0",
|
| 245 |
+
"triton": "3.7.0"
|
| 246 |
+
}
|
| 247 |
+
},
|
| 248 |
+
"output_directory": "/data/llmjp-pj/machuang/projects/mi_object_direction/codebase/sr_looped_pret_local/runs/synka-p2-15c46844396f00d2034b7a1e-26805-33c136b1/output",
|
| 249 |
+
"wandb_run_id": "synka-p2-15c46844396f00d2034b7a1e",
|
| 250 |
+
"slurm_job_id": "26805",
|
| 251 |
+
"gpu_processes": 8,
|
| 252 |
+
"dry_run": false,
|
| 253 |
+
"threads_per_rank": "8",
|
| 254 |
+
"new_experiment": true,
|
| 255 |
+
"resume_checkpoint": null,
|
| 256 |
+
"resume_payloads_verified": false,
|
| 257 |
+
"adapter_sha256": {
|
| 258 |
+
"launch_single_node.py": "a0061c18cdbb71368a85dac4ce24c9740d1d16cdcc234ee69ee5475ac827c447",
|
| 259 |
+
"python_module.py": "8a7bd07f472362b23df2d47d3abd2f0f924b2de94a4668641a28a9ae17ae615b",
|
| 260 |
+
"environment-lock.txt": "3f463f69cb2a7a5d1127d4421dff6d91be1b05071f9ed545c7d998eea255edd5",
|
| 261 |
+
"single_node.sbatch": "615eb73ea531fc23868a49b19639baa164fac5bb61fa8ff7649ef15be334def8",
|
| 262 |
+
"launch_contract.py": "31a111d06d47a07a814993972169d7feff4957b5cb6af42ae144846611cf6468",
|
| 263 |
+
"posttrain_cache.py": "30586a1982b01fd2af0dba12043377edea4e8f107492bf2b32a41a5a79523d75"
|
| 264 |
+
}
|
| 265 |
+
}
|