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": 74438.36910355715,
-      "required_pages_per_second_upper_estimate": 0.0064521291715746755
+      "required_token_positions_per_second": 74310.12162192195,
+      "required_pages_per_second_upper_estimate": 0.006441013004906807
     },
     {
       "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": 9304.796137944644,
-      "required_pages_per_second_upper_estimate": 0.0037118222985258673
+      "required_token_positions_per_second": 9288.765202740244,
+      "required_pages_per_second_upper_estimate": 0.0037054273187889916
     },
     {
       "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": 18609.59227588929,
-      "required_pages_per_second_upper_estimate": 0.004305001655389048
+      "required_token_positions_per_second": 18577.530405480487,
+      "required_pages_per_second_upper_estimate": 0.0042975847059396185
     },
     {
       "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": 27914.388413833934,
-      "required_pages_per_second_upper_estimate": 0.0016998397989251548
+      "required_token_positions_per_second": 27866.29560822073,
+      "required_pages_per_second_upper_estimate": 0.0016969111993867612
     },
     {
       "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": 27914.388413833934,
-      "required_pages_per_second_upper_estimate": 0.002148926892069692
+      "required_token_positions_per_second": 27866.29560822073,
+      "required_pages_per_second_upper_estimate": 0.002145224574764181
     },
     {
       "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": 27914.388413833934,
-      "required_pages_per_second_upper_estimate": 0.34660328065155066
+      "required_token_positions_per_second": 27866.29560822073,
+      "required_pages_per_second_upper_estimate": 0.34600612896210103
     }
   ],
   "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": 0.27010924582035045,
-    "producer_p95_seconds": 0.2977329280001868,
-    "gpu_update_seconds": 0.9904569496603438,
-    "producer_batches_per_second": 3.702205738877593,
-    "gpu_batches_per_second": 1.0096349976068408,
-    "producer_headroom": 3.6668754031437203,
+    "producer_batch_seconds": 0.12930864060132308,
+    "producer_p95_seconds": 0.33521719300188124,
+    "gpu_update_seconds": 0.992166321233012,
+    "producer_batches_per_second": 7.733435254981469,
+    "gpu_batches_per_second": 1.0078955298112244,
+    "producer_headroom": 7.6728540074286435,
     "sustainable": true,
-    "buffer_seconds": 60.41787392928097,
+    "buffer_seconds": 60.52214559521373,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346875.80738162855,
-    "upload_bytes_per_second": 5449378.02798872,
+    "generation_bytes_per_second": 333474.70806783985,
+    "upload_bytes_per_second": 5491044.59592118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.43147340857165,
+    "estimated_full_shards_per_hour": 71.55590945745847,
     "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": 0.9655942159988626,
-      "bytes": 7298211,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 0.8577407669999957,
       "bytes": 1078743,
@@ -562,6 +556,12 @@
       "bytes": 1103964,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1820991080021486,
+      "bytes": 11679771,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 336900, "loss": 0.6639048252254725, "lr": 0.0003, "grad_norm": 0.16959647781398787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49313718546, "tokens_per_second": 149526.03471198716, "data_wait_s": 1.549800072098151e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337000, "loss": 0.5864039976149797, "lr": 0.0003, "grad_norm": 0.17749098000646407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49328423295, "tokens_per_second": 148740.57350397322, "data_wait_s": 1.7457001376897097e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 337100, "loss": 0.5967775508761406, "lr": 0.0003, "grad_norm": 0.1694853529085336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49342909185, "tokens_per_second": 146831.1946399677, "data_wait_s": 1.6570000298088416e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 337200, "loss": 0.5765970721840858, "lr": 0.0003, "grad_norm": 0.1630461719034561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49357314524, "tokens_per_second": 146343.8726274883, "data_wait_s": 1.1293999705230817e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8924e14e513496ecd7c698c71e25c04fedef7aba63089871221579ed9e65160e
-size 471375
+oid sha256:0d1a893355483f6e18f3cf75195dfa9c70de33a2ca67860b69e7620c55a72e94
+size 473493