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": 72990.64592343297,
-      "required_pages_per_second_upper_estimate": 0.0063266441955423655
+      "required_token_positions_per_second": 74411.82034599382,
+      "required_pages_per_second_upper_estimate": 0.0064498279925562976
     },
     {
       "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": 9123.83074042912,
-      "required_pages_per_second_upper_estimate": 0.0036396324957831184
+      "required_token_positions_per_second": 9301.477543249228,
+      "required_pages_per_second_upper_estimate": 0.0037104984614844535
     },
     {
       "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": 18247.66148085824,
-      "required_pages_per_second_upper_estimate": 0.004221275335723053
+      "required_token_positions_per_second": 18602.955086498456,
+      "required_pages_per_second_upper_estimate": 0.004303466258434022
     },
     {
       "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": 27371.49222128736,
-      "required_pages_per_second_upper_estimate": 0.0016667802691552583
+      "required_token_positions_per_second": 27904.432629747687,
+      "required_pages_per_second_upper_estimate": 0.0016992335439082646
     },
     {
       "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": 27371.49222128736,
-      "required_pages_per_second_upper_estimate": 0.0021071332403346105
+      "required_token_positions_per_second": 27904.432629747687,
+      "required_pages_per_second_upper_estimate": 0.0021481604682513583
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27371.49222128736,
-      "required_pages_per_second_upper_estimate": 0.33986232689679724
+      "required_token_positions_per_second": 27904.432629747687,
+      "required_pages_per_second_upper_estimate": 0.3464796631330654
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.45052330407031604,
-    "producer_p95_seconds": 0.2875314709999657,
-    "gpu_update_seconds": 1.0101020352298364,
-    "producer_batches_per_second": 2.2196410062817162,
-    "gpu_batches_per_second": 0.989998995272257,
-    "producer_headroom": 2.242063897924764,
+    "producer_batch_seconds": 0.07077223542194133,
+    "producer_p95_seconds": 0.25897187699956703,
+    "gpu_update_seconds": 0.9908103263323722,
+    "producer_batches_per_second": 14.129834871514777,
+    "gpu_batches_per_second": 1.0092749070365916,
+    "producer_headroom": 13.99998630006809,
     "sustainable": true,
-    "buffer_seconds": 60.606122113790185,
+    "buffer_seconds": 60.439429906274704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303615.0132141588,
-    "upload_bytes_per_second": 5581702.388245194,
+    "generation_bytes_per_second": 303941.2244534273,
+    "upload_bytes_per_second": 5734255.004429556,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.14871403997968,
+    "estimated_full_shards_per_hour": 65.2187113781177,
     "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": 1.1758933869987231,
-      "bytes": 2380578,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.1564406300003611,
       "bytes": 5972956,
@@ -562,6 +556,12 @@
       "bytes": 1187973,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.1084424000000581,
+      "bytes": 13383772,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 376700, "loss": 0.583888977766037, "lr": 0.0003, "grad_norm": 0.16854099997862482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55137931768, "tokens_per_second": 150553.43137612595, "data_wait_s": 2.1908999769948423e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376800, "loss": 0.5738610681146383, "lr": 0.0003, "grad_norm": 0.16267895038316044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55152684511, "tokens_per_second": 149972.65571879316, "data_wait_s": 1.3511998986359686e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376900, "loss": 0.5763358268886805, "lr": 0.0003, "grad_norm": 0.1399091670930659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55167351142, "tokens_per_second": 148658.51848274763, "data_wait_s": 1.2592001439770684e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 377000, "loss": 0.5250621605664492, "lr": 0.0003, "grad_norm": 0.1611399413595201, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55182000710, "tokens_per_second": 149231.57059292105, "data_wait_s": 1.909200000227429e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f68d908309dbc824e01a5749f454a43210017ef8069a601dde5f14117905f90e
-size 467139
+oid sha256:3a79ab8c63dedd363c6e9ac9ff096ded07f625bfbfa03d4f803e186d8ef61214
+size 469257