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": 74887.30708661932,
-      "required_pages_per_second_upper_estimate": 0.0064910419781235885
+      "required_token_positions_per_second": 74887.08250471727,
+      "required_pages_per_second_upper_estimate": 0.006491022511933516
     },
     {
       "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": 9360.913385827414,
-      "required_pages_per_second_upper_estimate": 0.0037342083077339296
+      "required_token_positions_per_second": 9360.885313089659,
+      "required_pages_per_second_upper_estimate": 0.0037341971090991137
     },
     {
       "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": 18721.82677165483,
-      "required_pages_per_second_upper_estimate": 0.0043309651307247435
+      "required_token_positions_per_second": 18721.770626179317,
+      "required_pages_per_second_upper_estimate": 0.00433095214245711
     },
     {
       "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": 28082.740157482243,
-      "required_pages_per_second_upper_estimate": 0.001710091536840001
+      "required_token_positions_per_second": 28082.655939268974,
+      "required_pages_per_second_upper_estimate": 0.0017100864083926717
     },
     {
       "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": 28082.740157482243,
-      "required_pages_per_second_upper_estimate": 0.0021618870753231925
+      "required_token_positions_per_second": 28082.655939268974,
+      "required_pages_per_second_upper_estimate": 0.002161880591972004
     },
     {
       "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": 28082.740157482243,
-      "required_pages_per_second_upper_estimate": 0.3486936458706215
+      "required_token_positions_per_second": 28082.655939268974,
+      "required_pages_per_second_upper_estimate": 0.34869260016227294
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10384596087482123,
-    "producer_p95_seconds": 0.34032460500020534,
-    "gpu_update_seconds": 0.9845193113263054,
-    "producer_batches_per_second": 9.629647523849554,
-    "gpu_batches_per_second": 1.0157241087052316,
-    "producer_headroom": 9.480573948495426,
+    "producer_batch_seconds": 0.15211595453109794,
+    "producer_p95_seconds": 0.3268552970002929,
+    "gpu_update_seconds": 0.984522263841107,
+    "producer_batches_per_second": 6.573932386530594,
+    "gpu_batches_per_second": 1.0157210626182356,
+    "producer_headroom": 6.472182795525471,
     "sustainable": true,
-    "buffer_seconds": 60.055677990904634,
+    "buffer_seconds": 60.055858094307524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291858.3609035804,
-    "upload_bytes_per_second": 4772366.749063736,
+    "generation_bytes_per_second": 331871.3488988658,
+    "upload_bytes_per_second": 4775567.4506229935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.62601013498839,
+    "estimated_full_shards_per_hour": 71.21186590408783,
     "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": 1.0292037019971758,
-      "bytes": 5575102,
-      "pages": 18,
-      "shards": 2
-    },
     {
       "seconds": 1.288112906000606,
       "bytes": 5599143,
@@ -562,6 +556,12 @@
       "bytes": 2642711,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9964947309999843,
+      "bytes": 5697842,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 267600, "loss": 0.5919433161616325, "lr": 0.0003, "grad_norm": 0.15521812435714716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39168598739, "tokens_per_second": 147657.67878407598, "data_wait_s": 1.7366000975016505e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 267700, "loss": 0.5945382248610258, "lr": 0.0003, "grad_norm": 0.15191960258312273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39183308033, "tokens_per_second": 149902.31310225226, "data_wait_s": 1.5765999705763534e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 267800, "loss": 0.5631391536444426, "lr": 0.0003, "grad_norm": 0.14669065333577141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39197877179, "tokens_per_second": 148099.78003376513, "data_wait_s": 1.5671001165173948e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 267900, "loss": 0.6203124411404133, "lr": 0.0003, "grad_norm": 0.1908258995673714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39212530776, "tokens_per_second": 149008.2329960209, "data_wait_s": 1.1150998034281656e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d71a8568ddf9302922b14f7e6a6d79db806124bd6e11c30581892c9748cbff76
-size 655089
+oid sha256:d8394e24928a540c12252063e1935556ee415b592386d2e20a16aa055e2b2cd7
+size 657207