{
"cells": [
{
"cell_type": "markdown",
"id": "481c7ee8",
"metadata": {},
"source": [
"### Start by making sure you have your API key setup"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "a0d70ed1",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"\n",
"if os.getenv(\"A3_API_KEY\", \"\") == \"\":\n",
" raise ValueError(\n",
" \"A3_API_KEY is not set. Please set it in your environment variables or use `os.environ['A3_API_KEY'] = 'your-8-digit-student-id'`\"\n",
" )"
]
},
{
"cell_type": "markdown",
"id": "97b64b44",
"metadata": {},
"source": [
"### Do imports"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "10118323",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/herman/dev/cs336-teach/cs336-monorepo/assignments/scaling/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
" from .autonotebook import tqdm as notebook_tqdm\n"
]
}
],
"source": [
"import time\n",
"import rich\n",
"\n",
"from cs336_scaling.client import (\n",
" get_budget,\n",
" get_experiment,\n",
" list_experiments,\n",
" save_final_submission,\n",
" submit_experiment,\n",
")\n",
"from cs336_scaling.training.model.config import BasicTransformerConfig\n",
"from cs336_scaling.training.optimizer import AdamWConfig, WarmupCosineDecay\n",
"from cs336_scaling.training.training_config import TrainingConfig"
]
},
{
"cell_type": "markdown",
"id": "2d98fe94",
"metadata": {},
"source": [
"### 1. Start by checking your budget. It should be empty for now"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "4b7aa14f",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"BudgetSummary(used_seconds=0.0, remaining_seconds=43200.0, total_budget_seconds=43200.0)"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"get_budget()"
]
},
{
"cell_type": "markdown",
"id": "5b53793a",
"metadata": {},
"source": [
"### 2. Submit an experiment"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "76be1f80",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-06,\n",
" rope_theta=1000000,\n",
" tie_word_embeddings=False,\n",
" dtype='bfloat16',\n",
" vocab_size=32000\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(peak_value=0.0003, final_lr_frac=0.1, warmup_frac=0.05, init_value=0.0),\n",
" weight_decay=0.01,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-08,\n",
" eps_root=1e-08,\n",
" grad_clip_norm=1.0\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8388608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0\n",
")\n",
"\n"
],
"text/plain": [
"\u001b[1;35mTrainingConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33marchitecture_config\u001b[0m=\u001b[1;35mBasicTransformerConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mattention_bias\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mhead_dim\u001b[0m=\u001b[1;36m64\u001b[0m,\n",
" \u001b[33mhidden_size\u001b[0m=\u001b[1;36m448\u001b[0m,\n",
" \u001b[33mintermediate_size\u001b[0m=\u001b[1;36m1280\u001b[0m,\n",
" \u001b[33mnum_attention_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mnum_hidden_layers\u001b[0m=\u001b[1;36m9\u001b[0m,\n",
" \u001b[33mnum_key_value_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mrms_norm_eps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-06\u001b[0m,\n",
" \u001b[33mrope_theta\u001b[0m=\u001b[1;36m1000000\u001b[0m,\n",
" \u001b[33mtie_word_embeddings\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mdtype\u001b[0m=\u001b[32m'bfloat16'\u001b[0m,\n",
" \u001b[33mvocab_size\u001b[0m=\u001b[1;36m32000\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33moptimizer_config\u001b[0m=\u001b[1;35mAdamWConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mlr_scheduler\u001b[0m=\u001b[1;35mWarmupCosineDecay\u001b[0m\u001b[1m(\u001b[0m\u001b[33mpeak_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0003\u001b[0m, \u001b[33mfinal_lr_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.1\u001b[0m, \u001b[33mwarmup_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.05\u001b[0m, \u001b[33minit_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mweight_decay\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.01\u001b[0m,\n",
" \u001b[33mbeta1\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.9\u001b[0m,\n",
" \u001b[33mbeta2\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.95\u001b[0m,\n",
" \u001b[33meps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33meps_root\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33mgrad_clip_norm\u001b[0m=\u001b[1;36m1\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mtrain_batch_size\u001b[0m=\u001b[1;36m128\u001b[0m,\n",
" \u001b[33mval_batch_size\u001b[0m=\u001b[1;36m32\u001b[0m,\n",
" \u001b[33mn_evals\u001b[0m=\u001b[1;36m16\u001b[0m,\n",
" \u001b[33mtotal_train_tokens\u001b[0m=\u001b[1;36m8388608\u001b[0m,\n",
" \u001b[33mmax_runtime_seconds\u001b[0m=\u001b[1;36m60\u001b[0m\u001b[1;36m.0\u001b[0m,\n",
" \u001b[33mmodel_seed\u001b[0m=\u001b[1;36m0\u001b[0m\n",
"\u001b[1m)\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# define our experiment config\n",
"training_config = TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-6,\n",
" rope_theta=1_000_000,\n",
" tie_word_embeddings=False,\n",
" dtype=\"bfloat16\",\n",
" vocab_size=32_000,\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(\n",
" peak_value=3e-4,\n",
" final_lr_frac=0.1,\n",
" warmup_frac=0.05,\n",
" init_value=0.0,\n",
" ),\n",
" weight_decay=1e-2,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-8,\n",
" eps_root=1e-8,\n",
" grad_clip_norm=1.0,\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8_388_608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0,\n",
")\n",
"rich.print(training_config)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "18cd1d09",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"SubmitResponse(experiment_id=14, budget_summary=BudgetSummary(used_seconds=60.0, remaining_seconds=43140.0, total_budget_seconds=43200.0))"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# submit the experiment\n",
"submit_result = submit_experiment(training_config)\n",
"submit_result"
]
},
{
"cell_type": "markdown",
"id": "ee3eec4e",
"metadata": {},
"source": [
"#### You can also submit your config as a JSON-string\n",
"\n",
"The code below will crash, since you've already submitted the same experiment above."
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "64f224cb",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Experiment already exists for this training config (expected, continuing).\n"
]
}
],
"source": [
"try:\n",
" submit_result_json = submit_experiment(\n",
" {\n",
" \"architecture_config\": {\n",
" \"attention_bias\": False,\n",
" \"head_dim\": 64,\n",
" \"hidden_size\": 448,\n",
" \"intermediate_size\": 1280,\n",
" \"num_attention_heads\": 7,\n",
" \"num_hidden_layers\": 9,\n",
" \"num_key_value_heads\": 7,\n",
" \"rms_norm_eps\": 1e-06,\n",
" \"rope_theta\": 1000000,\n",
" \"tie_word_embeddings\": False,\n",
" \"dtype\": \"bfloat16\",\n",
" \"vocab_size\": 32000,\n",
" },\n",
" \"optimizer_config\": {\n",
" \"lr_scheduler\": {\n",
" \"peak_value\": 0.0003,\n",
" \"final_lr_frac\": 0.1,\n",
" \"warmup_frac\": 0.05,\n",
" \"init_value\": 0.0,\n",
" },\n",
" \"weight_decay\": 0.01,\n",
" \"beta1\": 0.9,\n",
" \"beta2\": 0.95,\n",
" \"eps\": 1e-08,\n",
" \"eps_root\": 1e-08,\n",
" \"grad_clip_norm\": 1.0,\n",
" },\n",
" \"train_batch_size\": 128,\n",
" \"val_batch_size\": 32,\n",
" \"n_evals\": 16,\n",
" \"total_train_tokens\": 8388608,\n",
" \"max_runtime_seconds\": 60.0,\n",
" \"model_seed\": 0,\n",
" }\n",
" )\n",
"except RuntimeError as e:\n",
" if \"409 Client Error: Conflict\" in str(e) and \"experiment already exists\" in str(e):\n",
" print(\n",
" \"Experiment already exists for this training config (expected, continuing).\"\n",
" )\n",
" submit_result_json = None\n",
" else:\n",
" raise"
]
},
{
"cell_type": "markdown",
"id": "fb3babe0",
"metadata": {},
"source": [
"### Query experiments"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "702d8273",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"[\n",
" ExperimentResponse(\n",
" experiment_id=14,\n",
" training_config=TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-06,\n",
" rope_theta=1000000,\n",
" tie_word_embeddings=False,\n",
" dtype='bfloat16',\n",
" vocab_size=32000\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(\n",
" peak_value=0.0003,\n",
" final_lr_frac=0.1,\n",
" warmup_frac=0.05,\n",
" init_value=0.0\n",
" ),\n",
" weight_decay=0.01,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-08,\n",
" eps_root=1e-08,\n",
" grad_clip_norm=1.0\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8388608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0\n",
" ),\n",
" status=QueuedExperimentStatus(\n",
" queued_at=datetime.datetime(2026, 4, 29, 23, 46, 36, 972380, tzinfo=TzInfo(0)),\n",
" status_type='queued'\n",
" )\n",
" )\n",
"]\n",
"\n"
],
"text/plain": [
"\u001b[1m[\u001b[0m\n",
" \u001b[1;35mExperimentResponse\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mexperiment_id\u001b[0m=\u001b[1;36m14\u001b[0m,\n",
" \u001b[33mtraining_config\u001b[0m=\u001b[1;35mTrainingConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33marchitecture_config\u001b[0m=\u001b[1;35mBasicTransformerConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mattention_bias\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mhead_dim\u001b[0m=\u001b[1;36m64\u001b[0m,\n",
" \u001b[33mhidden_size\u001b[0m=\u001b[1;36m448\u001b[0m,\n",
" \u001b[33mintermediate_size\u001b[0m=\u001b[1;36m1280\u001b[0m,\n",
" \u001b[33mnum_attention_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mnum_hidden_layers\u001b[0m=\u001b[1;36m9\u001b[0m,\n",
" \u001b[33mnum_key_value_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mrms_norm_eps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-06\u001b[0m,\n",
" \u001b[33mrope_theta\u001b[0m=\u001b[1;36m1000000\u001b[0m,\n",
" \u001b[33mtie_word_embeddings\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mdtype\u001b[0m=\u001b[32m'bfloat16'\u001b[0m,\n",
" \u001b[33mvocab_size\u001b[0m=\u001b[1;36m32000\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33moptimizer_config\u001b[0m=\u001b[1;35mAdamWConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mlr_scheduler\u001b[0m=\u001b[1;35mWarmupCosineDecay\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mpeak_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0003\u001b[0m,\n",
" \u001b[33mfinal_lr_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.1\u001b[0m,\n",
" \u001b[33mwarmup_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.05\u001b[0m,\n",
" \u001b[33minit_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mweight_decay\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.01\u001b[0m,\n",
" \u001b[33mbeta1\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.9\u001b[0m,\n",
" \u001b[33mbeta2\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.95\u001b[0m,\n",
" \u001b[33meps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33meps_root\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33mgrad_clip_norm\u001b[0m=\u001b[1;36m1\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mtrain_batch_size\u001b[0m=\u001b[1;36m128\u001b[0m,\n",
" \u001b[33mval_batch_size\u001b[0m=\u001b[1;36m32\u001b[0m,\n",
" \u001b[33mn_evals\u001b[0m=\u001b[1;36m16\u001b[0m,\n",
" \u001b[33mtotal_train_tokens\u001b[0m=\u001b[1;36m8388608\u001b[0m,\n",
" \u001b[33mmax_runtime_seconds\u001b[0m=\u001b[1;36m60\u001b[0m\u001b[1;36m.0\u001b[0m,\n",
" \u001b[33mmodel_seed\u001b[0m=\u001b[1;36m0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mstatus\u001b[0m=\u001b[1;35mQueuedExperimentStatus\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mqueued_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m46\u001b[0m, \u001b[1;36m36\u001b[0m, \u001b[1;36m972380\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mstatus_type\u001b[0m=\u001b[32m'queued'\u001b[0m\n",
" \u001b[1m)\u001b[0m\n",
" \u001b[1m)\u001b[0m\n",
"\u001b[1m]\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# the experiment is in queued state for now\n",
"rich.print(list_experiments())"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "5392619f",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"[\n",
" ExperimentResponse(\n",
" experiment_id=14,\n",
" training_config=TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-06,\n",
" rope_theta=1000000,\n",
" tie_word_embeddings=False,\n",
" dtype='bfloat16',\n",
" vocab_size=32000\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(\n",
" peak_value=0.0003,\n",
" final_lr_frac=0.1,\n",
" warmup_frac=0.05,\n",
" init_value=0.0\n",
" ),\n",
" weight_decay=0.01,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-08,\n",
" eps_root=1e-08,\n",
" grad_clip_norm=1.0\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8388608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0\n",
" ),\n",
" status=RunningExperimentStatus(\n",
" queued_at=datetime.datetime(2026, 4, 29, 23, 46, 36, 972380, tzinfo=TzInfo(0)),\n",
" dispatched_at=datetime.datetime(2026, 4, 29, 23, 46, 38, 244101, tzinfo=TzInfo(0)),\n",
" run_id='fc-01KQDT6BXQFMPGVWR3GTS3PVJ3',\n",
" val_losses=[],\n",
" status_type='running'\n",
" )\n",
" )\n",
"]\n",
"\n"
],
"text/plain": [
"\u001b[1m[\u001b[0m\n",
" \u001b[1;35mExperimentResponse\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mexperiment_id\u001b[0m=\u001b[1;36m14\u001b[0m,\n",
" \u001b[33mtraining_config\u001b[0m=\u001b[1;35mTrainingConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33marchitecture_config\u001b[0m=\u001b[1;35mBasicTransformerConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mattention_bias\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mhead_dim\u001b[0m=\u001b[1;36m64\u001b[0m,\n",
" \u001b[33mhidden_size\u001b[0m=\u001b[1;36m448\u001b[0m,\n",
" \u001b[33mintermediate_size\u001b[0m=\u001b[1;36m1280\u001b[0m,\n",
" \u001b[33mnum_attention_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mnum_hidden_layers\u001b[0m=\u001b[1;36m9\u001b[0m,\n",
" \u001b[33mnum_key_value_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mrms_norm_eps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-06\u001b[0m,\n",
" \u001b[33mrope_theta\u001b[0m=\u001b[1;36m1000000\u001b[0m,\n",
" \u001b[33mtie_word_embeddings\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mdtype\u001b[0m=\u001b[32m'bfloat16'\u001b[0m,\n",
" \u001b[33mvocab_size\u001b[0m=\u001b[1;36m32000\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33moptimizer_config\u001b[0m=\u001b[1;35mAdamWConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mlr_scheduler\u001b[0m=\u001b[1;35mWarmupCosineDecay\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mpeak_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0003\u001b[0m,\n",
" \u001b[33mfinal_lr_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.1\u001b[0m,\n",
" \u001b[33mwarmup_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.05\u001b[0m,\n",
" \u001b[33minit_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mweight_decay\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.01\u001b[0m,\n",
" \u001b[33mbeta1\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.9\u001b[0m,\n",
" \u001b[33mbeta2\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.95\u001b[0m,\n",
" \u001b[33meps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33meps_root\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33mgrad_clip_norm\u001b[0m=\u001b[1;36m1\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mtrain_batch_size\u001b[0m=\u001b[1;36m128\u001b[0m,\n",
" \u001b[33mval_batch_size\u001b[0m=\u001b[1;36m32\u001b[0m,\n",
" \u001b[33mn_evals\u001b[0m=\u001b[1;36m16\u001b[0m,\n",
" \u001b[33mtotal_train_tokens\u001b[0m=\u001b[1;36m8388608\u001b[0m,\n",
" \u001b[33mmax_runtime_seconds\u001b[0m=\u001b[1;36m60\u001b[0m\u001b[1;36m.0\u001b[0m,\n",
" \u001b[33mmodel_seed\u001b[0m=\u001b[1;36m0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mstatus\u001b[0m=\u001b[1;35mRunningExperimentStatus\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mqueued_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m46\u001b[0m, \u001b[1;36m36\u001b[0m, \u001b[1;36m972380\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mdispatched_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m46\u001b[0m, \u001b[1;36m38\u001b[0m, \u001b[1;36m244101\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mrun_id\u001b[0m=\u001b[32m'fc-01KQDT6BXQFMPGVWR3GTS3PVJ3'\u001b[0m,\n",
" \u001b[33mval_losses\u001b[0m=\u001b[1m[\u001b[0m\u001b[1m]\u001b[0m,\n",
" \u001b[33mstatus_type\u001b[0m=\u001b[32m'running'\u001b[0m\n",
" \u001b[1m)\u001b[0m\n",
" \u001b[1m)\u001b[0m\n",
"\u001b[1m]\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# after a few seconds, it will be in running state, as long as the queue is not full\n",
"time.sleep(10)\n",
"rich.print(list_experiments())"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "78ce7bec",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"ExperimentResponse(\n",
" experiment_id=14,\n",
" training_config=TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-06,\n",
" rope_theta=1000000,\n",
" tie_word_embeddings=False,\n",
" dtype='bfloat16',\n",
" vocab_size=32000\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(\n",
" peak_value=0.0003,\n",
" final_lr_frac=0.1,\n",
" warmup_frac=0.05,\n",
" init_value=0.0\n",
" ),\n",
" weight_decay=0.01,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-08,\n",
" eps_root=1e-08,\n",
" grad_clip_norm=1.0\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8388608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0\n",
" ),\n",
" status=CompletedExperimentStatus(\n",
" queued_at=datetime.datetime(2026, 4, 29, 23, 46, 36, 972380, tzinfo=TzInfo(0)),\n",
" dispatched_at=datetime.datetime(2026, 4, 29, 23, 46, 38, 244101, tzinfo=TzInfo(0)),\n",
" run_id='fc-01KQDT6BXQFMPGVWR3GTS3PVJ3',\n",
" used_runtime_seconds=10.18408550300228,\n",
" val_losses=[\n",
" 9.625,\n",
" 8.92578125,\n",
" 8.37890625,\n",
" 7.97265625,\n",
" 7.607421875,\n",
" 7.4453125,\n",
" 7.328125,\n",
" 7.291015625,\n",
" 7.25,\n",
" 7.22265625,\n",
" 7.208984375,\n",
" 7.19921875,\n",
" 7.193359375,\n",
" 7.189453125,\n",
" 7.189453125,\n",
" 7.189453125\n",
" ],\n",
" completed_at=datetime.datetime(2026, 4, 29, 23, 47, 25, 805998, tzinfo=TzInfo(0)),\n",
" status_type='completed'\n",
" )\n",
")\n",
"\n"
],
"text/plain": [
"\u001b[1;35mExperimentResponse\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mexperiment_id\u001b[0m=\u001b[1;36m14\u001b[0m,\n",
" \u001b[33mtraining_config\u001b[0m=\u001b[1;35mTrainingConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33marchitecture_config\u001b[0m=\u001b[1;35mBasicTransformerConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mattention_bias\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mhead_dim\u001b[0m=\u001b[1;36m64\u001b[0m,\n",
" \u001b[33mhidden_size\u001b[0m=\u001b[1;36m448\u001b[0m,\n",
" \u001b[33mintermediate_size\u001b[0m=\u001b[1;36m1280\u001b[0m,\n",
" \u001b[33mnum_attention_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mnum_hidden_layers\u001b[0m=\u001b[1;36m9\u001b[0m,\n",
" \u001b[33mnum_key_value_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mrms_norm_eps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-06\u001b[0m,\n",
" \u001b[33mrope_theta\u001b[0m=\u001b[1;36m1000000\u001b[0m,\n",
" \u001b[33mtie_word_embeddings\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mdtype\u001b[0m=\u001b[32m'bfloat16'\u001b[0m,\n",
" \u001b[33mvocab_size\u001b[0m=\u001b[1;36m32000\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33moptimizer_config\u001b[0m=\u001b[1;35mAdamWConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mlr_scheduler\u001b[0m=\u001b[1;35mWarmupCosineDecay\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mpeak_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0003\u001b[0m,\n",
" \u001b[33mfinal_lr_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.1\u001b[0m,\n",
" \u001b[33mwarmup_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.05\u001b[0m,\n",
" \u001b[33minit_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mweight_decay\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.01\u001b[0m,\n",
" \u001b[33mbeta1\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.9\u001b[0m,\n",
" \u001b[33mbeta2\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.95\u001b[0m,\n",
" \u001b[33meps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33meps_root\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33mgrad_clip_norm\u001b[0m=\u001b[1;36m1\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mtrain_batch_size\u001b[0m=\u001b[1;36m128\u001b[0m,\n",
" \u001b[33mval_batch_size\u001b[0m=\u001b[1;36m32\u001b[0m,\n",
" \u001b[33mn_evals\u001b[0m=\u001b[1;36m16\u001b[0m,\n",
" \u001b[33mtotal_train_tokens\u001b[0m=\u001b[1;36m8388608\u001b[0m,\n",
" \u001b[33mmax_runtime_seconds\u001b[0m=\u001b[1;36m60\u001b[0m\u001b[1;36m.0\u001b[0m,\n",
" \u001b[33mmodel_seed\u001b[0m=\u001b[1;36m0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mstatus\u001b[0m=\u001b[1;35mCompletedExperimentStatus\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mqueued_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m46\u001b[0m, \u001b[1;36m36\u001b[0m, \u001b[1;36m972380\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mdispatched_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m46\u001b[0m, \u001b[1;36m38\u001b[0m, \u001b[1;36m244101\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mrun_id\u001b[0m=\u001b[32m'fc-01KQDT6BXQFMPGVWR3GTS3PVJ3'\u001b[0m,\n",
" \u001b[33mused_runtime_seconds\u001b[0m=\u001b[1;36m10\u001b[0m\u001b[1;36m.18408550300228\u001b[0m,\n",
" \u001b[33mval_losses\u001b[0m=\u001b[1m[\u001b[0m\n",
" \u001b[1;36m9.625\u001b[0m,\n",
" \u001b[1;36m8.92578125\u001b[0m,\n",
" \u001b[1;36m8.37890625\u001b[0m,\n",
" \u001b[1;36m7.97265625\u001b[0m,\n",
" \u001b[1;36m7.607421875\u001b[0m,\n",
" \u001b[1;36m7.4453125\u001b[0m,\n",
" \u001b[1;36m7.328125\u001b[0m,\n",
" \u001b[1;36m7.291015625\u001b[0m,\n",
" \u001b[1;36m7.25\u001b[0m,\n",
" \u001b[1;36m7.22265625\u001b[0m,\n",
" \u001b[1;36m7.208984375\u001b[0m,\n",
" \u001b[1;36m7.19921875\u001b[0m,\n",
" \u001b[1;36m7.193359375\u001b[0m,\n",
" \u001b[1;36m7.189453125\u001b[0m,\n",
" \u001b[1;36m7.189453125\u001b[0m,\n",
" \u001b[1;36m7.189453125\u001b[0m\n",
" \u001b[1m]\u001b[0m,\n",
" \u001b[33mcompleted_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m47\u001b[0m, \u001b[1;36m25\u001b[0m, \u001b[1;36m805998\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m,\n",
" \u001b[33mstatus_type\u001b[0m=\u001b[32m'completed'\u001b[0m\n",
" \u001b[1m)\u001b[0m\n",
"\u001b[1m)\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"# finally, it will turn into finished state\n",
"time.sleep(60)\n",
"rich.print(get_experiment(submit_result.experiment_id))"
]
},
{
"cell_type": "markdown",
"id": "1f10110c",
"metadata": {},
"source": [
"### We are refunded the difference between the budget and the cost of the experiment once it finishes\n",
"\n",
"We reserved 60 seconds for our experiment, but "
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "f8e5d42c",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"BudgetSummary(used_seconds=10.18408550300228, remaining_seconds=43189.815914497, total_budget_seconds=43200.0)"
]
},
"execution_count": 10,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"get_budget()"
]
},
{
"cell_type": "markdown",
"id": "5020138f",
"metadata": {},
"source": [
"### Submit your final validation. This will run for 48 B200-hours"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "9ab4c6a0",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"FinalSubmissionResponse(\n",
" training_config=TrainingConfig(\n",
" architecture_config=BasicTransformerConfig(\n",
" attention_bias=False,\n",
" head_dim=64,\n",
" hidden_size=448,\n",
" intermediate_size=1280,\n",
" num_attention_heads=7,\n",
" num_hidden_layers=9,\n",
" num_key_value_heads=7,\n",
" rms_norm_eps=1e-06,\n",
" rope_theta=1000000,\n",
" tie_word_embeddings=False,\n",
" dtype='bfloat16',\n",
" vocab_size=32000\n",
" ),\n",
" optimizer_config=AdamWConfig(\n",
" lr_scheduler=WarmupCosineDecay(\n",
" peak_value=0.0003,\n",
" final_lr_frac=0.1,\n",
" warmup_frac=0.05,\n",
" init_value=0.0\n",
" ),\n",
" weight_decay=0.01,\n",
" beta1=0.9,\n",
" beta2=0.95,\n",
" eps=1e-08,\n",
" eps_root=1e-08,\n",
" grad_clip_norm=1.0\n",
" ),\n",
" train_batch_size=128,\n",
" val_batch_size=32,\n",
" n_evals=16,\n",
" total_train_tokens=8388608,\n",
" max_runtime_seconds=60.0,\n",
" model_seed=0\n",
" ),\n",
" predicted_final_loss=7.2,\n",
" submitted_at=datetime.datetime(2026, 4, 29, 23, 47, 47, 218799, tzinfo=TzInfo(0))\n",
")\n",
"\n"
],
"text/plain": [
"\u001b[1;35mFinalSubmissionResponse\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mtraining_config\u001b[0m=\u001b[1;35mTrainingConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33marchitecture_config\u001b[0m=\u001b[1;35mBasicTransformerConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mattention_bias\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mhead_dim\u001b[0m=\u001b[1;36m64\u001b[0m,\n",
" \u001b[33mhidden_size\u001b[0m=\u001b[1;36m448\u001b[0m,\n",
" \u001b[33mintermediate_size\u001b[0m=\u001b[1;36m1280\u001b[0m,\n",
" \u001b[33mnum_attention_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mnum_hidden_layers\u001b[0m=\u001b[1;36m9\u001b[0m,\n",
" \u001b[33mnum_key_value_heads\u001b[0m=\u001b[1;36m7\u001b[0m,\n",
" \u001b[33mrms_norm_eps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-06\u001b[0m,\n",
" \u001b[33mrope_theta\u001b[0m=\u001b[1;36m1000000\u001b[0m,\n",
" \u001b[33mtie_word_embeddings\u001b[0m=\u001b[3;91mFalse\u001b[0m,\n",
" \u001b[33mdtype\u001b[0m=\u001b[32m'bfloat16'\u001b[0m,\n",
" \u001b[33mvocab_size\u001b[0m=\u001b[1;36m32000\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33moptimizer_config\u001b[0m=\u001b[1;35mAdamWConfig\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mlr_scheduler\u001b[0m=\u001b[1;35mWarmupCosineDecay\u001b[0m\u001b[1m(\u001b[0m\n",
" \u001b[33mpeak_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0003\u001b[0m,\n",
" \u001b[33mfinal_lr_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.1\u001b[0m,\n",
" \u001b[33mwarmup_frac\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.05\u001b[0m,\n",
" \u001b[33minit_value\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mweight_decay\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.01\u001b[0m,\n",
" \u001b[33mbeta1\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.9\u001b[0m,\n",
" \u001b[33mbeta2\u001b[0m=\u001b[1;36m0\u001b[0m\u001b[1;36m.95\u001b[0m,\n",
" \u001b[33meps\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33meps_root\u001b[0m=\u001b[1;36m1e\u001b[0m\u001b[1;36m-08\u001b[0m,\n",
" \u001b[33mgrad_clip_norm\u001b[0m=\u001b[1;36m1\u001b[0m\u001b[1;36m.0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mtrain_batch_size\u001b[0m=\u001b[1;36m128\u001b[0m,\n",
" \u001b[33mval_batch_size\u001b[0m=\u001b[1;36m32\u001b[0m,\n",
" \u001b[33mn_evals\u001b[0m=\u001b[1;36m16\u001b[0m,\n",
" \u001b[33mtotal_train_tokens\u001b[0m=\u001b[1;36m8388608\u001b[0m,\n",
" \u001b[33mmax_runtime_seconds\u001b[0m=\u001b[1;36m60\u001b[0m\u001b[1;36m.0\u001b[0m,\n",
" \u001b[33mmodel_seed\u001b[0m=\u001b[1;36m0\u001b[0m\n",
" \u001b[1m)\u001b[0m,\n",
" \u001b[33mpredicted_final_loss\u001b[0m=\u001b[1;36m7\u001b[0m\u001b[1;36m.2\u001b[0m,\n",
" \u001b[33msubmitted_at\u001b[0m=\u001b[1;35mdatetime\u001b[0m\u001b[1;35m.datetime\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m2026\u001b[0m, \u001b[1;36m4\u001b[0m, \u001b[1;36m29\u001b[0m, \u001b[1;36m23\u001b[0m, \u001b[1;36m47\u001b[0m, \u001b[1;36m47\u001b[0m, \u001b[1;36m218799\u001b[0m, \u001b[33mtzinfo\u001b[0m=\u001b[1;35mTzInfo\u001b[0m\u001b[1m(\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1m)\u001b[0m\u001b[1m)\u001b[0m\n",
"\u001b[1m)\u001b[0m\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"rich.print(save_final_submission(training_config, 7.2))"
]
}
],
"metadata": {
"kernelspec": {
"display_name": ".venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}