Upload think-d12-1ep-65sh-r30 step 2500

experiments/think-d12-1ep-65sh-r30/base_checkpoints/meta_002500.json

@@ -0,0 +1,155 @@
+{
+  "step": 2500,
+  "experiment_id": "think-d12-1ep-65sh-r30",
+  "val_bpb": 1.1919329110762835,
+  "model_config": {
+    "sequence_len": 2048,
+    "vocab_size": 32768,
+    "n_layer": 12,
+    "n_head": 6,
+    "n_kv_head": 6,
+    "n_embd": 768,
+    "window_pattern": "L"
+  },
+  "user_config": {
+    "run": "think-d12-1ep-65sh-r30",
+    "wandb_run_id": "6465e19b",
+    "wandb_group": "think-d12-stopping-point",
+    "wandb_tags": "think-dataset,d12,ratio30,65-shards,a100,bf16,stopping-point",
+    "device_type": "cuda",
+    "fp8": false,
+    "fp8_recipe": "tensorwise",
+    "depth": 12,
+    "aspect_ratio": 64,
+    "head_dim": 128,
+    "max_seq_len": 2048,
+    "window_pattern": "L",
+    "num_iterations": -1,
+    "target_flops": -1.0,
+    "target_param_data_ratio": 30.0,
+    "device_batch_size": 16,
+    "total_batch_size": 524288,
+    "embedding_lr": 0.3,
+    "unembedding_lr": 0.008,
+    "weight_decay": 0.28,
+    "matrix_lr": 0.02,
+    "scalar_lr": 0.5,
+    "warmup_steps": 40,
+    "warmdown_ratio": 0.65,
+    "final_lr_frac": 0.05,
+    "resume_from_step": -1,
+    "pretokenized": true,
+    "data_dir": "/content/nanochat_cache/experiments/think-d12-1ep-65sh-r30/data",
+    "tokenizer_dir": "/content/nanochat_cache/experiments/think-d12-1ep-65sh-r30/tokenizer",
+    "pretokenized_dir": "/content/nanochat_cache/experiments/think-d12-1ep-65sh-r30/pretok",
+    "checkpoint_dir": "/content/nanochat_cache/experiments/think-d12-1ep-65sh-r30/base_checkpoints",
+    "experiment_id": "think-d12-1ep-65sh-r30",
+    "experiment_config": "/content/nanochat_cache/experiments/think-d12-1ep-65sh-r30/config.json",
+    "tokenizer_fingerprint": "db3bec0946e70097",
+    "git_commit_sha": "09b5f3b7b61a45f34d50a36e570b85f0259208bc",
+    "eval_every": 250,
+    "eval_tokens": 2097152,
+    "core_metric_every": 500,
+    "core_metric_max_per_task": 50,
+    "sample_every": -1,
+    "save_every": 500,
+    "model_tag": "think-d12-1ep-65sh-r30",
+    "experiment": {
+      "schema_version": 1,
+      "stage": "base",
+      "experiment_id": "think-d12-1ep-65sh-r30",
+      "dataset": {
+        "adapter": "parquet_shards",
+        "repo": "jbduran/think-dataset",
+        "revision": "main",
+        "validation_shard": 472,
+        "num_train_shards": 65,
+        "download_workers": 4
+      },
+      "tokenizer": {
+        "mode": "train",
+        "max_chars": 2000000000,
+        "doc_cap": 10000,
+        "vocab_size": 32768
+      },
+      "pretokenize": {
+        "enabled": true,
+        "target_tokens": 3402104832,
+        "slack": 1.03,
+        "require_no_wrap": true,
+        "val_tokens": 20971520,
+        "shard_tokens": 100000000,
+        "tokenizer_threads": 8
+      },
+      "training": {
+        "device_type": "cuda",
+        "depth": 12,
+        "aspect_ratio": 64,
+        "head_dim": 128,
+        "max_seq_len": 2048,
+        "scaling_params": 110100912,
+        "target_param_data_ratio": 30.0,
+        "window_pattern": "L",
+        "device_batch_size": 16,
+        "total_batch_size": 524288,
+        "embedding_lr": 0.3,
+        "unembedding_lr": 0.008,
+        "weight_decay": 0.28,
+        "matrix_lr": 0.02,
+        "scalar_lr": 0.5,
+        "warmup_steps": 40,
+        "warmdown_ratio": 0.65,
+        "final_lr_frac": 0.05,
+        "save_every": 500,
+        "eval_every": 250,
+        "eval_tokens": 2097152,
+        "core_metric_every": 500,
+        "core_metric_max_per_task": 50,
+        "sample_every": -1
+      },
+      "artifacts": {
+        "repo": "jbduran/think.nano"
+      },
+      "wandb": {
+        "entity": "jbduran-thinkingmachinesncsu",
+        "project": "think.nano",
+        "name": "think-d12-1ep-65sh-r30",
+        "group": "think-d12-stopping-point",
+        "tags": [
+          "think-dataset",
+          "d12",
+          "ratio30",
+          "65-shards",
+          "a100",
+          "bf16",
+          "stopping-point"
+        ]
+      },
+      "config_fingerprint": "35996219a51996ca",
+      "artifact_path": "experiments/think-d12-1ep-65sh-r30"
+    },
+    "stage": "base",
+    "base_experiment_id": "think-d12-1ep-65sh-r30",
+    "parent_experiment_id": null,
+    "parent_checkpoint_step": null,
+    "config_fingerprint": "35996219a51996ca"
+  },
+  "device_batch_size": 16,
+  "max_seq_len": 2048,
+  "total_batch_size": 524288,
+  "dataloader_state_dict": {
+    "file_idx": 13,
+    "pos": 10792769,
+    "epoch": 1,
+    "pq_idx": 13,
+    "rg_idx": 10792769
+  },
+  "loop_state": {
+    "min_val_bpb": 1.1919329110762835,
+    "smooth_train_loss": 3.4785363277537242,
+    "total_training_time": 6687.278959035873,
+    "stage_training_flops": 1162736943759360000,
+    "inherited_parent_flops": 0.0,
+    "cumulative_pipeline_training_flops": 1162736943759360000
+  }
+}