Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75548.31327078487,
-      "required_pages_per_second_upper_estimate": 0.0065483363188622015
+      "required_token_positions_per_second": 75582.98986103677,
+      "required_pages_per_second_upper_estimate": 0.006551341997818222
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9443.539158848109,
-      "required_pages_per_second_upper_estimate": 0.00376716896395728
+      "required_token_positions_per_second": 9447.873732629596,
+      "required_pages_per_second_upper_estimate": 0.0037688980902463682
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18887.078317696218,
-      "required_pages_per_second_upper_estimate": 0.00436919316757354
+      "required_token_positions_per_second": 18895.74746525919,
+      "required_pages_per_second_upper_estimate": 0.004371198622290448
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28330.61747654433,
-      "required_pages_per_second_upper_estimate": 0.0017251859650590977
+      "required_token_positions_per_second": 28343.621197888788,
+      "required_pages_per_second_upper_estimate": 0.001725977823463722
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28330.61747654433,
-      "required_pages_per_second_upper_estimate": 0.0021809693575129277
+      "required_token_positions_per_second": 28343.621197888788,
+      "required_pages_per_second_upper_estimate": 0.002181970419978629
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28330.61747654433,
-      "required_pages_per_second_upper_estimate": 0.3517714525813518
+      "required_token_positions_per_second": 28343.621197888788,
+      "required_pages_per_second_upper_estimate": 0.3519329152797942
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1473271315617808,
-    "producer_p95_seconds": 0.4110099169993191,
-    "gpu_update_seconds": 0.9759053089079515,
-    "producer_batches_per_second": 6.787616031067948,
-    "gpu_batches_per_second": 1.0246895788680674,
-    "producer_headroom": 6.62407051954793,
+    "producer_batch_seconds": 0.32690128728910395,
+    "producer_p95_seconds": 0.40769881100277416,
+    "gpu_update_seconds": 0.9754575749854926,
+    "producer_batches_per_second": 3.0590274155623716,
+    "gpu_batches_per_second": 1.0251599102245657,
+    "producer_headroom": 2.9839514645986096,
     "sustainable": true,
-    "buffer_seconds": 60.50612915229299,
+    "buffer_seconds": 60.47836964910054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441061.66913469526,
-    "upload_bytes_per_second": 3805839.5604445566,
+    "generation_bytes_per_second": 409274.0328113417,
+    "upload_bytes_per_second": 3754747.4225630225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.64156680613178,
+    "estimated_full_shards_per_hour": 87.82068002944172,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1450656299966795,
-      "bytes": 17577071,
-      "pages": 37,
-      "shards": 5
-    },
     {
       "seconds": 1.0943805140013865,
       "bytes": 891607,
@@ -562,6 +556,12 @@
       "bytes": 873310,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.2830225369980326,
+      "bytes": 12398676,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 36100, "loss": 0.613822378218174, "lr": 0.0003, "grad_norm": 0.1500501413705502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5282567214, "tokens_per_second": 149343.03274959428, "data_wait_s": 1.2067001080140471e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 36200, "loss": 0.7056152671575546, "lr": 0.0003, "grad_norm": 0.1365660897291415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5297376839, "tokens_per_second": 150446.74409384117, "data_wait_s": 1.2477001291699708e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 36300, "loss": 0.6558767966926098, "lr": 0.0003, "grad_norm": 0.13010661947009525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5311989777, "tokens_per_second": 148919.61666417075, "data_wait_s": 1.0694995580706745e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 36400, "loss": 0.6693796012550592, "lr": 0.0003, "grad_norm": 0.15807813006973326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5326521022, "tokens_per_second": 147688.64795002664, "data_wait_s": 1.2843003787565976e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:267885570cfc442e978f8b3b03188cc555bda3d874468128d222b285ff6a2952
-size 768706
+oid sha256:946bf8df2f203971469e0459625b59e14fda9ee2b632ad062bd9c860e29f7641
+size 770824