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": 75062.90721793451,
-      "required_pages_per_second_upper_estimate": 0.006506262552451526
+      "required_token_positions_per_second": 75871.42411205679,
+      "required_pages_per_second_upper_estimate": 0.006576342747666711
     },
     {
       "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": 9382.863402241814,
-      "required_pages_per_second_upper_estimate": 0.003742964497463624
+      "required_token_positions_per_second": 9483.928014007099,
+      "required_pages_per_second_upper_estimate": 0.0037832806821473984
     },
     {
       "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": 18765.726804483627,
-      "required_pages_per_second_upper_estimate": 0.004341120630705496
+      "required_token_positions_per_second": 18967.856028014197,
+      "required_pages_per_second_upper_estimate": 0.004387879669216458
     },
     {
       "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": 28148.59020672544,
-      "required_pages_per_second_upper_estimate": 0.0017141014593504038
+      "required_token_positions_per_second": 28451.784042021292,
+      "required_pages_per_second_upper_estimate": 0.001732564373184806
     },
     {
       "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": 28148.59020672544,
-      "required_pages_per_second_upper_estimate": 0.0021669563943985367
+      "required_token_positions_per_second": 28451.784042021292,
+      "required_pages_per_second_upper_estimate": 0.0021902970951338647
     },
     {
       "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": 28148.59020672544,
-      "required_pages_per_second_upper_estimate": 0.3495112830962842
+      "required_token_positions_per_second": 28451.784042021292,
+      "required_pages_per_second_upper_estimate": 0.3532759358061673
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1569631200236472,
-    "producer_p95_seconds": 0.43900408799527213,
-    "gpu_update_seconds": 0.9822161535249522,
-    "producer_batches_per_second": 6.37092330892343,
-    "gpu_batches_per_second": 1.0181058379168635,
-    "producer_headroom": 6.257623786893232,
+    "producer_batch_seconds": 0.09921103007025067,
+    "producer_p95_seconds": 0.4002200670001912,
+    "gpu_update_seconds": 0.9717492568889824,
+    "producer_batches_per_second": 10.07952441670958,
+    "gpu_batches_per_second": 1.0290720501309785,
+    "producer_headroom": 9.794770361731889,
     "sustainable": true,
-    "buffer_seconds": 60.897401518547035,
+    "buffer_seconds": 60.248453927116906,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319710.1647681754,
+    "generation_bytes_per_second": 319520.0545524263,
     "upload_bytes_per_second": 5798627.87923205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.60235888751933,
+    "estimated_full_shards_per_hour": 68.5615656607589,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 233200, "loss": 0.5668853744864464, "lr": 0.0003, "grad_norm": 0.14522512206479846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34134025710, "tokens_per_second": 149567.2520319809, "data_wait_s": 1.1172000085934997e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233300, "loss": 0.5669607650488615, "lr": 0.0003, "grad_norm": 0.13534628347548794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34148706901, "tokens_per_second": 148219.70475623052, "data_wait_s": 1.6645994037389755e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233400, "loss": 0.5881534330546856, "lr": 0.0003, "grad_norm": 0.15923444566418937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34163539739, "tokens_per_second": 151573.37001466786, "data_wait_s": 9.046998457051814e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 233500, "loss": 0.4985401816666126, "lr": 0.0003, "grad_norm": 0.13233641011653927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34178233929, "tokens_per_second": 147326.7281481473, "data_wait_s": 1.691799843683839e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b7e8c2d152ea625e12295fdb02bc1671daf1f737812b99470bb5cada63f20f5
-size 818958
+oid sha256:49b988f1534fd42a8737a431900268639fff02dfaab621e5b5cbb82f61e20a86
+size 821076