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": 74704.74614179804,
-      "required_pages_per_second_upper_estimate": 0.006475218058122964
+      "required_token_positions_per_second": 74711.44616106735,
+      "required_pages_per_second_upper_estimate": 0.006475798798812184
     },
     {
       "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": 9338.093267724755,
-      "required_pages_per_second_upper_estimate": 0.003725105021431608
+      "required_token_positions_per_second": 9338.930770133418,
+      "required_pages_per_second_upper_estimate": 0.0037254391136642005
     },
     {
       "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": 18676.18653544951,
-      "required_pages_per_second_upper_estimate": 0.0043204070653193654
+      "required_token_positions_per_second": 18677.861540266837,
+      "required_pages_per_second_upper_estimate": 0.004320794548204777
     },
     {
       "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": 28014.27980317426,
-      "required_pages_per_second_upper_estimate": 0.0017059226604463646
+      "required_token_positions_per_second": 28016.79231040025,
+      "required_pages_per_second_upper_estimate": 0.0017060756589543208
     },
     {
       "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": 28014.27980317426,
-      "required_pages_per_second_upper_estimate": 0.0021566168077381737
+      "required_token_positions_per_second": 28016.79231040025,
+      "required_pages_per_second_upper_estimate": 0.0021568102274994957
     },
     {
       "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": 28014.27980317426,
-      "required_pages_per_second_upper_estimate": 0.34784359739218323
+      "required_token_positions_per_second": 28016.79231040025,
+      "required_pages_per_second_upper_estimate": 0.34787479432310925
     }
   ],
   "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": 0.11489136274994394,
-    "producer_p95_seconds": 0.33646183100063354,
-    "gpu_update_seconds": 0.9869252464904431,
-    "producer_batches_per_second": 8.703874478157742,
-    "gpu_batches_per_second": 1.0132479674180506,
-    "producer_headroom": 8.590073464777705,
+    "producer_batch_seconds": 0.13734019103887363,
+    "producer_p95_seconds": 0.3250483730007545,
+    "gpu_update_seconds": 0.986836740397888,
+    "producer_batches_per_second": 7.281189813671904,
+    "gpu_batches_per_second": 1.0133388422453795,
+    "producer_headroom": 7.185345621942288,
     "sustainable": true,
-    "buffer_seconds": 60.20244003591703,
+    "buffer_seconds": 60.19704116427117,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327915.4920931405,
-    "upload_bytes_per_second": 5070623.427974382,
+    "generation_bytes_per_second": 327854.1026954605,
+    "upload_bytes_per_second": 5035934.569192295,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.36303112121259,
+    "estimated_full_shards_per_hour": 70.3498583855425,
     "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": 1.0578431400026602,
-      "bytes": 6390274,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1717372180028178,
       "bytes": 4762409,
@@ -562,6 +556,12 @@
       "bytes": 10018327,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.856085086998064,
+      "bytes": 7579977,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -245,3 +245,4 @@
 {"step": 261600, "loss": 0.6089760884642601, "lr": 0.0003, "grad_norm": 0.12999718473535907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38291041851, "tokens_per_second": 149931.84472369438, "data_wait_s": 9.758998203324154e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 261700, "loss": 0.5810473840683699, "lr": 0.0003, "grad_norm": 0.14975278904965478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38305751328, "tokens_per_second": 149156.38515814557, "data_wait_s": 1.4618999557569623e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261800, "loss": 0.5959047731012106, "lr": 0.0003, "grad_norm": 0.15352184300430943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38320504086, "tokens_per_second": 149531.34447158725, "data_wait_s": 1.666000025579706e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 261900, "loss": 0.585339205339551, "lr": 0.0003, "grad_norm": 0.1740283848217525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38335035150, "tokens_per_second": 148937.061681626, "data_wait_s": 1.7112000932684168e-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 sha256:2579da3cdee670264d8346c21842f2fca2c7675eba80da2419b11c679d9df226
-size 527226
+oid sha256:3b68be0c784b0eef7d5cf57da9da58f453319d453686c0fab9eab4d264511421
+size 529344