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": 75702.43321991837,
-      "required_pages_per_second_upper_estimate": 0.006561695045439653
+      "required_token_positions_per_second": 75945.39485631048,
+      "required_pages_per_second_upper_estimate": 0.006582754344301484
     },
     {
       "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": 9462.804152489796,
-      "required_pages_per_second_upper_estimate": 0.003774854057958272
+      "required_token_positions_per_second": 9493.17435703881,
+      "required_pages_per_second_upper_estimate": 0.003786969186627896
     },
     {
       "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": 18925.608304979593,
-      "required_pages_per_second_upper_estimate": 0.004378106402026933
+      "required_token_positions_per_second": 18986.34871407762,
+      "required_pages_per_second_upper_estimate": 0.004392157626677072
     },
     {
       "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": 28388.41245746939,
-      "required_pages_per_second_upper_estimate": 0.0017287053761706034
+      "required_token_positions_per_second": 28479.52307111643,
+      "required_pages_per_second_upper_estimate": 0.001734253534521262
     },
     {
       "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": 28388.41245746939,
-      "required_pages_per_second_upper_estimate": 0.0021854185751312856
+      "required_token_positions_per_second": 28479.52307111643,
+      "required_pages_per_second_upper_estimate": 0.0021924325223801554
     },
     {
       "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": 28388.41245746939,
-      "required_pages_per_second_upper_estimate": 0.3524890728170827
+      "required_token_positions_per_second": 28479.52307111643,
+      "required_pages_per_second_upper_estimate": 0.3536203617109705
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12066974485969695,
-    "producer_p95_seconds": 0.3551416080008494,
-    "gpu_update_seconds": 0.9739184972538126,
-    "producer_batches_per_second": 8.287081415169169,
-    "gpu_batches_per_second": 1.0267799644628686,
-    "producer_headroom": 8.070941878481557,
+    "producer_batch_seconds": 0.14992873260933948,
+    "producer_p95_seconds": 0.35971107899968047,
+    "gpu_update_seconds": 0.970802774012752,
+    "producer_batches_per_second": 6.66983561186795,
+    "gpu_batches_per_second": 1.0300753425606348,
+    "producer_headroom": 6.475094914210446,
     "sustainable": true,
-    "buffer_seconds": 60.38294682973638,
+    "buffer_seconds": 60.18977198879062,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348951.40511792904,
-    "upload_bytes_per_second": 5100191.613075302,
+    "generation_bytes_per_second": 349756.1616309234,
+    "upload_bytes_per_second": 5063000.7891543815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.87684836533931,
+    "estimated_full_shards_per_hour": 75.04953037925507,
     "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": 1.2115873000002466,
-      "bytes": 10372111,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1448558080010116,
       "bytes": 6260573,
@@ -562,6 +556,12 @@
       "bytes": 1121142,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2539431289987988,
+      "bytes": 7556763,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -388,3 +388,4 @@
 {"step": 275900, "loss": 0.5654884856194258, "lr": 0.0003, "grad_norm": 0.1431217299136518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40383055030, "tokens_per_second": 147731.86951555742, "data_wait_s": 1.160100509878248e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276000, "loss": 0.6485577970743179, "lr": 0.0003, "grad_norm": 0.1465950070699948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40397759181, "tokens_per_second": 150855.9904269503, "data_wait_s": 1.0570998711045831e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276100, "loss": 0.6443565767258406, "lr": 0.0003, "grad_norm": 0.1798039695456819, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40412682300, "tokens_per_second": 152806.60512173743, "data_wait_s": 1.4905999705661088e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276200, "loss": 0.6205026488751173, "lr": 0.0003, "grad_norm": 0.1742606156034186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40427451945, "tokens_per_second": 127623.37690401972, "data_wait_s": 1.6697995306458324e-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 sha256:97005cb46b47b8f3babd248dd2d7246167be3d606926d50b0617845929f407a2
-size 832449
+oid sha256:5405f9555674770fa8a81bfa0c0dc0886c3d0071be7a837f0408df1160aba272
+size 834567