Add rig training logs: 177 runs across six studies, full resolution (part 2)

.gitattributes

@@ -134,3 +134,62 @@ batch-sweep-500M/500m-5tpp-bs256-lr2e-8-s1338/training.riglog filter=lfs diff=lf
 batch-sweep-500M/500m-5tpp-bs256-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
 batch-sweep-500M/500m-5tpp-bs256-lr2e-8-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
 batch-sweep-60M/60m-5tpp-bs128-lr2e-10-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-10-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-10-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs128-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-10-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-10-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-10-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-6-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-6-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-6-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs256-lr2e-9-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-10-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-6-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-7-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-8-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+batch-sweep-60M/60m-5tpp-bs32-lr2e-9-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+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-bs128-lr2e-10-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": 4.231022298336029,
+      "perplexity": 68.78751760617332,
+      "scored_tokens": 1048576,
+      "seconds": 0.11371587699977681
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1226.5817067241096,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 62.016062223003246,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.4254490790190175,
+    "final_validation_seconds": 0.11371587699977681,
+    "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": 0.2787685697100249,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3593723.0063139293,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 99.85941846802598,
+    "train_compile_seconds": 31.417907166003715,
+    "train_loss": 4.179998874664307,
+    "train_seconds": 83.37609534000512,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.231022298336029,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-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": 4.231022298336029,
+      "perplexity": 68.78751760617332,
+      "scored_tokens": 1048576,
+      "seconds": 0.11371587699977681
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1226.5817067241096,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 62.016062223003246,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.4254490790190175,
+    "final_validation_seconds": 0.11371587699977681,
+    "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": 0.2787685697100249,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3593723.0063139293,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 99.85941846802598,
+    "train_compile_seconds": 31.417907166003715,
+    "train_loss": 4.179998874664307,
+    "train_seconds": 83.37609534000512,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.231022298336029,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.231022298336029,68.78751760617332,0.11371587699977681,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-10-s1338/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e457af3ed771dcf138f2bca03cbc4c37a43401a74065653fafe2bd6c358abc1e
+size 272824

batch-sweep-60M/60m-5tpp-bs128-lr2e-10-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": 4.234494388103485,
+      "perplexity": 69.02676915325046,
+      "scored_tokens": 1048576,
+      "seconds": 0.11434176698094234
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1205.1196077005113,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 57.62204757099971,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.784824261005269,
+    "final_validation_seconds": 0.11434176698094234,
+    "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": 0.27389081993193437,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3530841.8801711923,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 99.80507420498179,
+    "train_compile_seconds": 35.39820237297681,
+    "train_loss": 4.222934722900391,
+    "train_seconds": 84.86094879600569,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.234494388103485,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-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": 4.234494388103485,
+      "perplexity": 69.02676915325046,
+      "scored_tokens": 1048576,
+      "seconds": 0.11434176698094234
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1205.1196077005113,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 57.62204757099971,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.784824261005269,
+    "final_validation_seconds": 0.11434176698094234,
+    "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": 0.27389081993193437,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3530841.8801711923,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 99.80507420498179,
+    "train_compile_seconds": 35.39820237297681,
+    "train_loss": 4.222934722900391,
+    "train_seconds": 84.86094879600569,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.234494388103485,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.234494388103485,69.02676915325046,0.11434176698094234,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-10-s1339/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adc6952fc772f08ca554883470ffe406e6a3b4c908afd8da02d749269040778f
+size 272824

batch-sweep-60M/60m-5tpp-bs128-lr2e-10-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": 4.22570413351059,
+      "perplexity": 68.42266528193188,
+      "scored_tokens": 1048576,
+      "seconds": 0.11953467698185705
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1230.2294350520417,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 57.58028576700599,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.514255159010645,
+    "final_validation_seconds": 0.11953467698185705,
+    "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": 0.27959759887546404,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3604410.370344397,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 94.90107854802045,
+    "train_compile_seconds": 30.806537622003816,
+    "train_loss": 4.224009037017822,
+    "train_seconds": 83.1288785719953,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.22570413351059,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-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": 4.22570413351059,
+      "perplexity": 68.42266528193188,
+      "scored_tokens": 1048576,
+      "seconds": 0.11953467698185705
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0009765625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0009765625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1230.2294350520417,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0009765625,
+    "diagnostic_compile_seconds": 57.58028576700599,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.514255159010645,
+    "final_validation_seconds": 0.11953467698185705,
+    "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": 0.27959759887546404,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3604410.370344397,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 94.90107854802045,
+    "train_compile_seconds": 30.806537622003816,
+    "train_loss": 4.224009037017822,
+    "train_seconds": 83.1288785719953,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.22570413351059,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-10-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.22570413351059,68.42266528193188,0.11953467698185705,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ae7005acc639a8c2dfb5ba44126c2aafd0f819d276bd8a305e1234b4ca1d717
+size 272824

batch-sweep-60M/60m-5tpp-bs128-lr2e-6-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": 4.095564603805542,
+      "perplexity": 60.07324716821631,
+      "scored_tokens": 1048576,
+      "seconds": 0.12321466801222414
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1220.6404960310078,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 62.52575223299209,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.569469711976126,
+    "final_validation_seconds": 0.12321466801222414,
+    "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": 0.2774182945525018,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3576316.0407313583,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 98.82152125096763,
+    "train_compile_seconds": 29.72629930599942,
+    "train_loss": 4.049431800842285,
+    "train_seconds": 83.78191093500936,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.095564603805542,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-6-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": 4.095564603805542,
+      "perplexity": 60.07324716821631,
+      "scored_tokens": 1048576,
+      "seconds": 0.12321466801222414
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1220.6404960310078,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 62.52575223299209,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.569469711976126,
+    "final_validation_seconds": 0.12321466801222414,
+    "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": 0.2774182945525018,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3576316.0407313583,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 98.82152125096763,
+    "train_compile_seconds": 29.72629930599942,
+    "train_loss": 4.049431800842285,
+    "train_seconds": 83.78191093500936,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.095564603805542,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-6-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.095564603805542,60.07324716821631,0.12321466801222414,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1338/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce9c382973c6d6d3b7cee14c7ef6dad4213e6eabaa350832adac55e2cb24941c
+size 272824

batch-sweep-60M/60m-5tpp-bs128-lr2e-6-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": 4.123899519443512,
+      "perplexity": 61.799762378212606,
+      "scored_tokens": 1048576,
+      "seconds": 0.12320758801070042
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1231.1065312499722,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 61.87786437702016,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.20147384799202,
+    "final_validation_seconds": 0.12320758801070042,
+    "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": 0.2797969389204482,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3606980.146795468,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 98.26842717602267,
+    "train_compile_seconds": 30.18908895101049,
+    "train_loss": 4.1150898933410645,
+    "train_seconds": 83.06965378398309,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.123899519443512,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-6-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": 4.123899519443512,
+      "perplexity": 61.799762378212606,
+      "scored_tokens": 1048576,
+      "seconds": 0.12320758801070042
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1231.1065312499722,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 61.87786437702016,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.20147384799202,
+    "final_validation_seconds": 0.12320758801070042,
+    "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": 0.2797969389204482,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3606980.146795468,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 98.26842717602267,
+    "train_compile_seconds": 30.18908895101049,
+    "train_loss": 4.1150898933410645,
+    "train_seconds": 83.06965378398309,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.123899519443512,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-6-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.123899519443512,61.799762378212606,0.12320758801070042,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-6-s1339/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9977da206c34c79f0e1e332fa671cb72ed85db234bbcb240db39f8c44af536a
+size 272824

batch-sweep-60M/60m-5tpp-bs128-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": 4.1288798451423645,
+      "perplexity": 62.108313026564716,
+      "scored_tokens": 1048576,
+      "seconds": 0.11513088998617604
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1229.273374905453,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 62.80066886998247,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.637855462002335,
+    "final_validation_seconds": 0.11513088998617604,
+    "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": 0.279380312478512,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3601609.2399138846,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 100.527059791988,
+    "train_compile_seconds": 31.088535460003186,
+    "train_loss": 4.132540702819824,
+    "train_seconds": 83.19353156900615,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.1288798451423645,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-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": 4.1288798451423645,
+      "perplexity": 62.108313026564716,
+      "scored_tokens": 1048576,
+      "seconds": 0.11513088998617604
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.015625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.015625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1229.273374905453,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.015625,
+    "diagnostic_compile_seconds": 62.80066886998247,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.637855462002335,
+    "final_validation_seconds": 0.11513088998617604,
+    "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": 0.279380312478512,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3601609.2399138846,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 100.527059791988,
+    "train_compile_seconds": 31.088535460003186,
+    "train_loss": 4.132540702819824,
+    "train_seconds": 83.19353156900615,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.1288798451423645,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-6-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.1288798451423645,62.108313026564716,0.11513088998617604,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47e320ae401697ebd9106108c0800e33410061c9876f888bdd7dbe9224bde653
+size 272824

batch-sweep-60M/60m-5tpp-bs128-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": 4.161057770252228,
+      "perplexity": 64.13933140730494,
+      "scored_tokens": 1048576,
+      "seconds": 0.11261016898788512
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1231.595172992706,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.25604649700108,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.325576535979053,
+    "final_validation_seconds": 0.11261016898788512,
+    "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": 0.2799079938619786,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3608411.8028058917,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.09768495999742,
+    "train_compile_seconds": 29.51606192701729,
+    "train_loss": 4.113357067108154,
+    "train_seconds": 83.03669547001482,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.161057770252228,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-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": 4.161057770252228,
+      "perplexity": 64.13933140730494,
+      "scored_tokens": 1048576,
+      "seconds": 0.11261016898788512
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1231.595172992706,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.25604649700108,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.325576535979053,
+    "final_validation_seconds": 0.11261016898788512,
+    "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": 0.2799079938619786,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3608411.8028058917,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.09768495999742,
+    "train_compile_seconds": 29.51606192701729,
+    "train_loss": 4.113357067108154,
+    "train_seconds": 83.03669547001482,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.161057770252228,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-7-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.161057770252228,64.13933140730494,0.11261016898788512,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1338/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c62bb909c29d1688f7508672beaea71259585a7c0a90bbc06b07332370643fec
+size 272824

batch-sweep-60M/60m-5tpp-bs128-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": 4.128543436527252,
+      "perplexity": 62.087422769020804,
+      "scored_tokens": 1048576,
+      "seconds": 0.11667717900127172
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1230.4733700894026,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.1994503029855,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.6018604100099765,
+    "final_validation_seconds": 0.11667717900127172,
+    "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": 0.2796530386566824,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3605125.0679067397,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.69660827299231,
+    "train_compile_seconds": 29.895297559996834,
+    "train_loss": 4.116345405578613,
+    "train_seconds": 83.11239869799465,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.128543436527252,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-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": 4.128543436527252,
+      "perplexity": 62.087422769020804,
+      "scored_tokens": 1048576,
+      "seconds": 0.11667717900127172
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1230.4733700894026,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.1994503029855,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.6018604100099765,
+    "final_validation_seconds": 0.11667717900127172,
+    "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": 0.2796530386566824,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3605125.0679067397,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.69660827299231,
+    "train_compile_seconds": 29.895297559996834,
+    "train_loss": 4.116345405578613,
+    "train_seconds": 83.11239869799465,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.128543436527252,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-7-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.128543436527252,62.087422769020804,0.11667717900127172,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-7-s1339/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13d677c5b9648f74728e7eac02e3577b6f4584d69439eaaf462f99415161ea29
+size 272824

batch-sweep-60M/60m-5tpp-bs128-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": 4.150980353355408,
+      "perplexity": 63.49621852610368,
+      "scored_tokens": 1048576,
+      "seconds": 0.12393446799251251
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1207.7172178718833,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.46536698698765,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.814937568997266,
+    "final_validation_seconds": 0.12393446799251251,
+    "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": 0.2744811858799735,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3538452.5361781423,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.77288727398263,
+    "train_compile_seconds": 29.49258271799772,
+    "train_loss": 4.149023056030273,
+    "train_seconds": 84.67842621498858,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.150980353355408,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-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": 4.150980353355408,
+      "perplexity": 63.49621852610368,
+      "scored_tokens": 1048576,
+      "seconds": 0.12393446799251251
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.0078125,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.0078125,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1207.7172178718833,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.0078125,
+    "diagnostic_compile_seconds": 61.46536698698765,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.814937568997266,
+    "final_validation_seconds": 0.12393446799251251,
+    "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": 0.2744811858799735,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3538452.5361781423,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 97.77288727398263,
+    "train_compile_seconds": 29.49258271799772,
+    "train_loss": 4.149023056030273,
+    "train_seconds": 84.67842621498858,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.150980353355408,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-7-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.150980353355408,63.49621852610368,0.12393446799251251,true

batch-sweep-60M/60m-5tpp-bs128-lr2e-8-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:040e5450dc9cbc67c105355097c00598335115b2943f726bf1890a53865fa051
+size 272824

batch-sweep-60M/60m-5tpp-bs128-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": 4.00784233212471,
+      "perplexity": 55.02801020912352,
+      "scored_tokens": 1048576,
+      "seconds": 0.11403720901580527
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1224.1597782331883,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 58.88308360200608,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.524559412006056,
+    "final_validation_seconds": 0.11403720901580527,
+    "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": 0.2782181314166337,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3586627.074514395,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 95.66653872502502,
+    "train_compile_seconds": 30.258895711012883,
+    "train_loss": 3.962064743041992,
+    "train_seconds": 83.5410500659782,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.00784233212471,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-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": 4.00784233212471,
+      "perplexity": 55.02801020912352,
+      "scored_tokens": 1048576,
+      "seconds": 0.11403720901580527
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "239f36c4cb1e574c1b4adbbe01104d4d17828e8a85abffb21ba4f0037f0c08ef",
+      "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": 128,
+        "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": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 229,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_multiplier": 1.0,
+          "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": 128,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 1024,
+          "steps": 2286,
+          "tokens_per_parameter": 5.000660099848113,
+          "train_tokens": 299630592
+        }
+      },
+      "schema_version": 2,
+      "sha256": "64473be07689abbf926027ef96fa7b679206510e29bae19403d94890c4fa1a7c"
+    },
+    "loss_backend": "tiled",
+    "vocab_tile_size": 2048
+  },
+  "metrics": {
+    "achieved_tflops": 1224.1597782331883,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 58.88308360200608,
+    "diagnostic_point_count": 230,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 102267593322135552,
+    "eval_compile_seconds": 6.524559412006056,
+    "final_validation_seconds": 0.11403720901580527,
+    "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": 0.2782181314166337,
+    "model_tier": "60m",
+    "parameter_count": 59918208,
+    "parameters": 59918208,
+    "tokens_per_parameter": 5.000660099848113,
+    "tokens_per_second": 3586627.074514395,
+    "tokens_processed": 299630592,
+    "total_compile_seconds": 95.66653872502502,
+    "train_compile_seconds": 30.258895711012883,
+    "train_loss": 3.962064743041992,
+    "train_seconds": 83.5410500659782,
+    "training_data_epochs": 0.037928020133706226,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 2286,
+    "training_token_budget": 299630592,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 4.00784233212471,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "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-bs128-lr2e-8-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+2286,299630592,fineweb,fineweb,1048576,4.00784233212471,55.02801020912352,0.11403720901580527,true