update slot resume pointers for step 1385

slots/13/checkpoint-672/config.json

@@ -1,24 +0,0 @@
-{
-  "architectures": [
-    "TwinyForCausalLM"
-  ],
-  "attention_dropout": 0.0,
-  "dtype": "float32",
-  "hidden_dropout": 0.0,
-  "hidden_size": 768,
-  "initializer_range": 0.02,
-  "intermediate_size": 3072,
-  "max_position_embeddings": 128,
-  "model_type": "twiny",
-  "neftune_alpha": 0.0,
-  "num_attention_heads": 12,
-  "num_hidden_layers": 3,
-  "num_key_value_heads": 3,
-  "qk_norm": true,
-  "rezero_init": 1.0,
-  "rms_norm_eps": 1e-06,
-  "rope_theta": 10000.0,
-  "transformers_version": "5.0.0",
-  "use_cache": false,
-  "vocab_size": 32000
-}

slots/13/checkpoint-672/trainer_state.json

@@ -1,272 +0,0 @@
-{
-  "best_global_step": null,
-  "best_metric": null,
-  "best_model_checkpoint": null,
-  "epoch": 0.089841824243589,
-  "eval_steps": 500,
-  "global_step": 672,
-  "is_hyper_param_search": false,
-  "is_local_process_zero": true,
-  "is_world_process_zero": true,
-  "log_history": [
-    {
-      "epoch": 0.0001336931908386741,
-      "grad_norm": Infinity,
-      "learning_rate": 5e-05,
-      "loss": 150.77503967285156,
-      "step": 1
-    },
-    {
-      "epoch": 0.002673863816773482,
-      "grad_norm": 65.5747299194336,
-      "learning_rate": 4.999995392022967e-05,
-      "loss": 65.57097424958882,
-      "step": 20
-    },
-    {
-      "epoch": 0.005347727633546964,
-      "grad_norm": 29.183452606201172,
-      "learning_rate": 4.999976672145381e-05,
-      "loss": 12.781006622314454,
-      "step": 40
-    },
-    {
-      "epoch": 0.008021591450320446,
-      "grad_norm": 22.14234161376953,
-      "learning_rate": 4.999943552476422e-05,
-      "loss": 7.058085632324219,
-      "step": 60
-    },
-    {
-      "epoch": 0.010695455267093928,
-      "grad_norm": 25.116792678833008,
-      "learning_rate": 4.999896033206858e-05,
-      "loss": 5.17547492980957,
-      "step": 80
-    },
-    {
-      "epoch": 0.01336931908386741,
-      "grad_norm": 17.95206642150879,
-      "learning_rate": 4.999834114610398e-05,
-      "loss": 3.6641620635986327,
-      "step": 100
-    },
-    {
-      "epoch": 0.01604318290064089,
-      "grad_norm": 13.571253776550293,
-      "learning_rate": 4.999757797043691e-05,
-      "loss": 2.9684438705444336,
-      "step": 120
-    },
-    {
-      "epoch": 0.018717046717414372,
-      "grad_norm": 11.740418434143066,
-      "learning_rate": 4.999667080946324e-05,
-      "loss": 2.658764457702637,
-      "step": 140
-    },
-    {
-      "epoch": 0.021390910534187856,
-      "grad_norm": 8.911559104919434,
-      "learning_rate": 4.999561966840821e-05,
-      "loss": 2.0916315078735352,
-      "step": 160
-    },
-    {
-      "epoch": 0.024064774350961337,
-      "grad_norm": 7.613410472869873,
-      "learning_rate": 4.9994424553326335e-05,
-      "loss": 1.730122184753418,
-      "step": 180
-    },
-    {
-      "epoch": 0.02673863816773482,
-      "grad_norm": 10.990550994873047,
-      "learning_rate": 4.999308547110146e-05,
-      "loss": 1.689904022216797,
-      "step": 200
-    },
-    {
-      "epoch": 0.029412501984508302,
-      "grad_norm": 13.414973258972168,
-      "learning_rate": 4.999160242944665e-05,
-      "loss": 1.331161880493164,
-      "step": 220
-    },
-    {
-      "epoch": 0.03208636580128178,
-      "grad_norm": 9.434646606445312,
-      "learning_rate": 4.998997543690418e-05,
-      "loss": 1.285347843170166,
-      "step": 240
-    },
-    {
-      "epoch": 0.03476022961805526,
-      "grad_norm": 15.048566818237305,
-      "learning_rate": 4.998820450284549e-05,
-      "loss": 1.1024582862854004,
-      "step": 260
-    },
-    {
-      "epoch": 0.037434093434828744,
-      "grad_norm": 7.001116752624512,
-      "learning_rate": 4.99862896374711e-05,
-      "loss": 1.106878662109375,
-      "step": 280
-    },
-    {
-      "epoch": 0.04010795725160223,
-      "grad_norm": 10.976016998291016,
-      "learning_rate": 4.998423085181056e-05,
-      "loss": 1.0516901969909669,
-      "step": 300
-    },
-    {
-      "epoch": 0.04278182106837571,
-      "grad_norm": 10.106513977050781,
-      "learning_rate": 4.998202815772245e-05,
-      "loss": 0.8515177726745605,
-      "step": 320
-    },
-    {
-      "epoch": 0.04545568488514919,
-      "grad_norm": 3.9024946689605713,
-      "learning_rate": 4.9979681567894195e-05,
-      "loss": 0.851468849182129,
-      "step": 340
-    },
-    {
-      "epoch": 0.048129548701922674,
-      "grad_norm": 7.649858474731445,
-      "learning_rate": 4.997719109584209e-05,
-      "loss": 0.8077478408813477,
-      "step": 360
-    },
-    {
-      "epoch": 0.050803412518696155,
-      "grad_norm": 11.689629554748535,
-      "learning_rate": 4.997455675591119e-05,
-      "loss": 0.6951910495758057,
-      "step": 380
-    },
-    {
-      "epoch": 0.05347727633546964,
-      "grad_norm": 12.950127601623535,
-      "learning_rate": 4.9971778563275204e-05,
-      "loss": 0.6042420387268066,
-      "step": 400
-    },
-    {
-      "epoch": 0.05615114015224312,
-      "grad_norm": 6.282811641693115,
-      "learning_rate": 4.9968856533936436e-05,
-      "loss": 0.6595364093780518,
-      "step": 420
-    },
-    {
-      "epoch": 0.058825003969016604,
-      "grad_norm": 3.553151845932007,
-      "learning_rate": 4.99657906847257e-05,
-      "loss": 0.71713285446167,
-      "step": 440
-    },
-    {
-      "epoch": 0.061498867785790085,
-      "grad_norm": 5.715928554534912,
-      "learning_rate": 4.996258103330218e-05,
-      "loss": 0.5080134391784668,
-      "step": 460
-    },
-    {
-      "epoch": 0.06417273160256357,
-      "grad_norm": 8.924579620361328,
-      "learning_rate": 4.995922759815339e-05,
-      "loss": 0.4877929210662842,
-      "step": 480
-    },
-    {
-      "epoch": 0.06684659541933705,
-      "grad_norm": 4.2473649978637695,
-      "learning_rate": 4.995573039859501e-05,
-      "loss": 0.5352929592132568,
-      "step": 500
-    },
-    {
-      "epoch": 0.06952045923611053,
-      "grad_norm": 5.880599021911621,
-      "learning_rate": 4.995208945477081e-05,
-      "loss": 0.520254373550415,
-      "step": 520
-    },
-    {
-      "epoch": 0.07219432305288401,
-      "grad_norm": 8.42399787902832,
-      "learning_rate": 4.994830478765251e-05,
-      "loss": 0.5289288520812988,
-      "step": 540
-    },
-    {
-      "epoch": 0.07486818686965749,
-      "grad_norm": 5.284440040588379,
-      "learning_rate": 4.9944376419039684e-05,
-      "loss": 0.4858880043029785,
-      "step": 560
-    },
-    {
-      "epoch": 0.07754205068643098,
-      "grad_norm": 4.662903308868408,
-      "learning_rate": 4.994030437155961e-05,
-      "loss": 0.4253716468811035,
-      "step": 580
-    },
-    {
-      "epoch": 0.08021591450320446,
-      "grad_norm": 2.9051289558410645,
-      "learning_rate": 4.993608866866718e-05,
-      "loss": 0.32270302772521975,
-      "step": 600
-    },
-    {
-      "epoch": 0.08288977831997794,
-      "grad_norm": 2.804189920425415,
-      "learning_rate": 4.993172933464471e-05,
-      "loss": 0.329010009765625,
-      "step": 620
-    },
-    {
-      "epoch": 0.08556364213675142,
-      "grad_norm": 1.455569863319397,
-      "learning_rate": 4.9927226394601815e-05,
-      "loss": 0.35200178623199463,
-      "step": 640
-    },
-    {
-      "epoch": 0.0882375059535249,
-      "grad_norm": 11.18322467803955,
-      "learning_rate": 4.992257987447532e-05,
-      "loss": 0.30103607177734376,
-      "step": 660
-    }
-  ],
-  "logging_steps": 20,
-  "max_steps": 26180,
-  "num_input_tokens_seen": 0,
-  "num_train_epochs": 4,
-  "save_steps": 1000000000,
-  "stateful_callbacks": {
-    "TrainerControl": {
-      "args": {
-        "should_epoch_stop": false,
-        "should_evaluate": false,
-        "should_log": false,
-        "should_save": true,
-        "should_training_stop": false
-      },
-      "attributes": {}
-    }
-  },
-  "total_flos": 2577328313204736.0,
-  "train_batch_size": 1,
-  "trial_name": null,
-  "trial_params": null
-}

slots/13/latest.json

@@ -1 +1 @@
-{"worker_id": "slot:13", "checkpoint": "slots/13/checkpoint-672", "step": 672, "updated_at": 1776679628}
+{"worker_id": "slot:13", "checkpoint": "slots/13/checkpoint-1385", "step": 1385, "updated_at": 1776682706}