Add rig training logs: 177 runs across six studies, full resolution (part 3)

.gitattributes

@@ -193,3 +193,50 @@ batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lf
 batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
 batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
 batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-10-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-10-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-10-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-6-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-6-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-6-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs64-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs128-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs256-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-7-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-8-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+lr-batch-sweep-125M/125m-5tpp-bs64-lr2e-9-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text

batch-sweep-60M/60m-5tpp-bs512-lr2e-6-s1339/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.460958361625671,
+      "perplexity": 235.3228407129802,
+      "scored_tokens": 4194304,
+      "seconds": 0.41062728301039897
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.03125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.864309241445,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 55.9644927130139,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.425732080009766,
+    "final_validation_seconds": 0.41062728301039897,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31451461573669204,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4054540.3363465653,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.20995718400809,
+    "train_compile_seconds": 27.819732390984427,
+    "train_loss": 5.450981616973877,
+    "train_seconds": 73.83536064898362,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.460958361625671,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-6-s1339/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.460958361625671,
+      "perplexity": 235.3228407129802,
+      "scored_tokens": 4194304,
+      "seconds": 0.41062728301039897
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.03125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.864309241445,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 55.9644927130139,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.425732080009766,
+    "final_validation_seconds": 0.41062728301039897,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31451461573669204,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4054540.3363465653,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.20995718400809,
+    "train_compile_seconds": 27.819732390984427,
+    "train_loss": 5.450981616973877,
+    "train_seconds": 73.83536064898362,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.460958361625671,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-6-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.460958361625671,235.3228407129802,0.41062728301039897,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1337/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.322399616241455,
+      "perplexity": 204.87491378400054,
+      "scored_tokens": 4194304,
+      "seconds": 0.42063087900169194
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1382.2561720529945,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 56.02680520899594,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.010083241009852,
+    "final_validation_seconds": 0.42063087900169194,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3141491300120442,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4049828.7059841016,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 92.69887297399691,
+    "train_compile_seconds": 27.661984523991123,
+    "train_loss": 5.302313804626465,
+    "train_seconds": 73.92126179501065,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.322399616241455,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1337/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.322399616241455,
+      "perplexity": 204.87491378400054,
+      "scored_tokens": 4194304,
+      "seconds": 0.42063087900169194
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1382.2561720529945,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 56.02680520899594,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.010083241009852,
+    "final_validation_seconds": 0.42063087900169194,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3141491300120442,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4049828.7059841016,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 92.69887297399691,
+    "train_compile_seconds": 27.661984523991123,
+    "train_loss": 5.302313804626465,
+    "train_seconds": 73.92126179501065,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.322399616241455,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.322399616241455,204.87491378400054,0.42063087900169194,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1338/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.34379643201828,
+      "perplexity": 209.30581915802543,
+      "scored_tokens": 4194304,
+      "seconds": 0.43600330100161955
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1363.4246614226945,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 60.228304310003296,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.48511767500895,
+    "final_validation_seconds": 0.43600330100161955,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3098692412324306,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3994654.8576992634,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 96.713636622997,
+    "train_compile_seconds": 27.000214637984755,
+    "train_loss": 5.338647842407227,
+    "train_seconds": 74.94225625600666,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.34379643201828,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1338/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.34379643201828,
+      "perplexity": 209.30581915802543,
+      "scored_tokens": 4194304,
+      "seconds": 0.43600330100161955
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1363.4246614226945,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 60.228304310003296,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.48511767500895,
+    "final_validation_seconds": 0.43600330100161955,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3098692412324306,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3994654.8576992634,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 96.713636622997,
+    "train_compile_seconds": 27.000214637984755,
+    "train_loss": 5.338647842407227,
+    "train_seconds": 74.94225625600666,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.34379643201828,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.34379643201828,209.30581915802543,0.43600330100161955,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1339/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.223019599914551,
+      "perplexity": 185.49345527979634,
+      "scored_tokens": 4194304,
+      "seconds": 0.4288240910100285
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1344.763268570482,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 55.81962695502443,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.311239109985763,
+    "final_validation_seconds": 0.4288240910100285,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3056280155842005,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3939979.4321200172,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 92.52819242299302,
+    "train_compile_seconds": 27.39732635798282,
+    "train_loss": 5.221643447875977,
+    "train_seconds": 75.9822362420091,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.223019599914551,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1339/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.223019599914551,
+      "perplexity": 185.49345527979634,
+      "scored_tokens": 4194304,
+      "seconds": 0.4288240910100285
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1344.763268570482,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 55.81962695502443,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.311239109985763,
+    "final_validation_seconds": 0.4288240910100285,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3056280155842005,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3939979.4321200172,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 92.52819242299302,
+    "train_compile_seconds": 27.39732635798282,
+    "train_loss": 5.221643447875977,
+    "train_seconds": 75.9822362420091,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.223019599914551,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-7-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.223019599914551,185.49345527979634,0.4288240910100285,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1337/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.341884195804596,
+      "perplexity": 208.90595942587223,
+      "scored_tokens": 4194304,
+      "seconds": 0.4429857379873283
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1356.7598127989045,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 56.21652633999474,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.699034475022927,
+    "final_validation_seconds": 0.4429857379873283,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3083545029088419,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3975127.728196507,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 94.05379035303486,
+    "train_compile_seconds": 28.138229538017185,
+    "train_loss": 5.310612678527832,
+    "train_seconds": 75.3103971669916,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.341884195804596,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1337/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.341884195804596,
+      "perplexity": 208.90595942587223,
+      "scored_tokens": 4194304,
+      "seconds": 0.4429857379873283
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1356.7598127989045,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 56.21652633999474,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.699034475022927,
+    "final_validation_seconds": 0.4429857379873283,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3083545029088419,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3975127.728196507,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 94.05379035303486,
+    "train_compile_seconds": 28.138229538017185,
+    "train_loss": 5.310612678527832,
+    "train_seconds": 75.3103971669916,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.341884195804596,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.341884195804596,208.90595942587223,0.4429857379873283,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1338/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.181924104690552,
+      "perplexity": 178.0250204412524,
+      "scored_tokens": 4194304,
+      "seconds": 0.4242735180014279
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.2889057202904,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 55.31145549900248,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.539940550021129,
+    "final_validation_seconds": 0.4242735180014279,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3143838422091569,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4052854.4797415365,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.02705095303827,
+    "train_compile_seconds": 28.175654904014664,
+    "train_loss": 5.17849063873291,
+    "train_seconds": 73.86607377501787,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.181924104690552,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1338/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.181924104690552,
+      "perplexity": 178.0250204412524,
+      "scored_tokens": 4194304,
+      "seconds": 0.4242735180014279
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.2889057202904,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 55.31145549900248,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.539940550021129,
+    "final_validation_seconds": 0.4242735180014279,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.3143838422091569,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4052854.4797415365,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.02705095303827,
+    "train_compile_seconds": 28.175654904014664,
+    "train_loss": 5.17849063873291,
+    "train_seconds": 73.86607377501787,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.181924104690552,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.181924104690552,178.0250204412524,0.4242735180014279,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1339/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.217997610569,
+      "perplexity": 184.56424432040083,
+      "scored_tokens": 4194304,
+      "seconds": 0.4158219499804545
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.881224560212,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 55.70425781101221,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.562512876989786,
+    "final_validation_seconds": 0.4158219499804545,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31451846012732093,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4054589.8960048244,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 95.73574358801125,
+    "train_compile_seconds": 30.46897290000925,
+    "train_loss": 5.217545509338379,
+    "train_seconds": 73.83445815197774,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.217997610569,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1339/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.217997610569,
+      "perplexity": 184.56424432040083,
+      "scored_tokens": 4194304,
+      "seconds": 0.4158219499804545
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1383.881224560212,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 55.70425781101221,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.562512876989786,
+    "final_validation_seconds": 0.4158219499804545,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31451846012732093,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4054589.8960048244,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 95.73574358801125,
+    "train_compile_seconds": 30.46897290000925,
+    "train_loss": 5.217545509338379,
+    "train_seconds": 73.83445815197774,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.217997610569,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-8-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.217997610569,184.56424432040083,0.4158219499804545,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1337/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.298588216304779,
+      "perplexity": 200.0541772879699,
+      "scored_tokens": 4194304,
+      "seconds": 0.41527405998203903
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1382.8013549102852,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 58.170241445011925,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.443252083001425,
+    "final_validation_seconds": 0.41527405998203903,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.314273035206883,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4051426.020020463,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 94.62185434199637,
+    "train_compile_seconds": 27.008360813983018,
+    "train_loss": 5.2749786376953125,
+    "train_seconds": 73.89211761998013,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.298588216304779,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1337/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.298588216304779,
+      "perplexity": 200.0541772879699,
+      "scored_tokens": 4194304,
+      "seconds": 0.41527405998203903
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1382.8013549102852,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 58.170241445011925,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.443252083001425,
+    "final_validation_seconds": 0.41527405998203903,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.314273035206883,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4051426.020020463,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 94.62185434199637,
+    "train_compile_seconds": 27.008360813983018,
+    "train_loss": 5.2749786376953125,
+    "train_seconds": 73.89211761998013,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.298588216304779,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.298588216304779,200.0541772879699,0.41527405998203903,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1338/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.610827386379242,
+      "perplexity": 273.37032734557306,
+      "scored_tokens": 4194304,
+      "seconds": 0.4116269589867443
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1358.4621183159854,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 55.512101172003895,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.259877471020445,
+    "final_validation_seconds": 0.4116269589867443,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.30874139052636035,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3980115.2593711303,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.47292305802694,
+    "train_compile_seconds": 28.700944415002596,
+    "train_loss": 5.606378555297852,
+    "train_seconds": 75.21602478600107,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.610827386379242,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1338/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.610827386379242,
+      "perplexity": 273.37032734557306,
+      "scored_tokens": 4194304,
+      "seconds": 0.4116269589867443
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1358.4621183159854,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 55.512101172003895,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.259877471020445,
+    "final_validation_seconds": 0.4116269589867443,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.30874139052636035,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 3980115.2593711303,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 93.47292305802694,
+    "train_compile_seconds": 28.700944415002596,
+    "train_loss": 5.606378555297852,
+    "train_seconds": 75.21602478600107,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.610827386379242,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.610827386379242,273.37032734557306,0.4116269589867443,true

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1339/metrics.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.csv",
+    "training_curve": "training.csv",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.322995722293854,
+      "perplexity": 204.99707736770014,
+      "scored_tokens": 4194304,
+      "seconds": 0.41537890999461524
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1384.0606298826538,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 55.89432832199964,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.284253924008226,
+    "final_validation_seconds": 0.41537890999461524,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31455923406423947,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4055115.5299933143,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 98.10837958700722,
+    "train_compile_seconds": 32.92979734099936,
+    "train_loss": 5.319055557250977,
+    "train_seconds": 73.82488754900987,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.322995722293854,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1339/result.json

@@ -0,0 +1,238 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 384,
+      "heads": 6,
+      "layers": 12,
+      "mlp_activation": "gelu",
+      "mlp_mult": 4,
+      "normalization": "rms_norm",
+      "parameterization": "complete_d_p",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "60m",
+      "vocab_size": 50304
+    },
+    "model_id": "reference-gpt-v3-family",
+    "sequence_length": 1024,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 5.322995722293854,
+      "perplexity": 204.99707736770014,
+      "scored_tokens": 4194304,
+      "seconds": 0.41537890999461524
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "95bc05cbf51b0f72ad8508ff3fd06150efd4c9fb645177f2678c63050ddcb857",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "overrides": {
+        "study_batch_size": 512,
+        "tokens_per_parameter_micros": 5000000
+      },
+      "path": "config.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "val_every": 0,
+          "val_probe_batches": 8
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "loss_backend": "tiled",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10
+        },
+        "model": {
+          "d_model": 384,
+          "heads": 6,
+          "layers": 12,
+          "mlp_activation": "gelu",
+          "mlp_mult": 4,
+          "normalization": "rms_norm",
+          "parameterization": "complete_d_p",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "60m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 0.5,
+            "beta1": 0.6000000000000001,
+            "beta2": 0.7999999999999998,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 2.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.001953125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 57,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 4.0,
+          "data_multiplier": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.0,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "name": "complete_d_p",
+          "width_multiplier": 1.0
+        },
+        "training": {
+          "batch_size": 512,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 571,
+          "tokens_per_parameter": 4.996285069139584,
+          "train_tokens": 299368448
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1384.0606298826538,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.001953125,
+    "diagnostic_compile_seconds": 55.89432832199964,
+    "diagnostic_point_count": 59,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102178120362098688,
+    "eval_compile_seconds": 9.284253924008226,
+    "final_validation_seconds": 0.41537890999461524,
+    "flop_accounting": {
+      "by_site": {
+        "dot_general": 291521691648,
+        "tpu_flash_causal_attention_bwd_dkv": 19327352832,
+        "tpu_flash_causal_attention_bwd_dq": 19327352832,
+        "tpu_flash_causal_attention_fwd": 19327352832
+      },
+      "elementwise_per_sequence": 1149350130,
+      "matmul_per_sequence": 349503750144,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 341312256,
+    "mfu_estimate": 0.31455923406423947,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 4.996285069139584,
+    "tokens_per_second": 4055115.5299933143,
+    "tokens_processed": 299368448,
+    "total_compile_seconds": 98.10837958700722,
+    "train_compile_seconds": 32.92979734099936,
+    "train_loss": 5.319055557250977,
+    "train_seconds": 73.82488754900987,
+    "training_data_epochs": 0.037894837263947954,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 571,
+    "training_token_budget": 299368448,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 5.322995722293854,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 4194304
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

batch-sweep-60M/60m-5tpp-bs512-lr2e-9-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+571,299368448,fineweb,fineweb,4194304,5.322995722293854,204.99707736770014,0.41537890999461524,true