Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75390.81684050769,
-      "required_pages_per_second_upper_estimate": 0.005215711639822152
+      "required_token_positions_per_second": 77009.66686826822,
+      "required_pages_per_second_upper_estimate": 0.0053277074409922
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9423.85210506346,
-      "required_pages_per_second_upper_estimate": 0.0039455700839801065
+      "required_token_positions_per_second": 9626.208358533528,
+      "required_pages_per_second_upper_estimate": 0.004030292421628933
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18847.70421012692,
-      "required_pages_per_second_upper_estimate": 0.0039254797395844275
+      "required_token_positions_per_second": 19252.416717067055,
+      "required_pages_per_second_upper_estimate": 0.004009770681793522
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28271.556315190384,
-      "required_pages_per_second_upper_estimate": 0.001694836632928688
+      "required_token_positions_per_second": 28878.625075600583,
+      "required_pages_per_second_upper_estimate": 0.001731229478174954
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28271.556315190384,
-      "required_pages_per_second_upper_estimate": 0.002197989150071898
+      "required_token_positions_per_second": 28878.625075600583,
+      "required_pages_per_second_upper_estimate": 0.0022451860759805107
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28271.556315190384,
-      "required_pages_per_second_upper_estimate": 0.3626371687791381
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28878.625075600583,
+      "required_pages_per_second_upper_estimate": 0.3704239950180293
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09489289803897805,
-    "producer_p95_seconds": 0.2201381439990655,
-    "gpu_update_seconds": 0.9779440400012451,
-    "producer_batches_per_second": 10.538196436884473,
-    "gpu_batches_per_second": 1.022553396816782,
-    "producer_headroom": 10.305766397813528,
+    "producer_batch_seconds": 0.12358969301567413,
+    "producer_p95_seconds": 0.22226372200020705,
+    "gpu_update_seconds": 0.9573862996462276,
+    "producer_batches_per_second": 8.091289618084707,
+    "gpu_batches_per_second": 1.0445104555700442,
+    "producer_headroom": 7.746489826824055,
     "sustainable": true,
-    "buffer_seconds": 60.6325304800772,
+    "buffer_seconds": 60.31533687771233,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468501.08164653654,
-    "upload_bytes_per_second": 5652535.764722333,
+    "generation_bytes_per_second": 489017.4438563139,
+    "upload_bytes_per_second": 5735639.290832505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.52942597434112,
+    "estimated_full_shards_per_hour": 104.93175970809042,
     "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": 1.077067022999472,
-      "bytes": 1093074,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.176304970000274,
       "bytes": 7496411,
@@ -562,6 +556,12 @@
       "bytes": 1138564,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0472652070020558,
+      "bytes": 7166131,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -188,3 +188,4 @@
 {"step": 703800, "loss": 0.5478715300559998, "lr": 8e-05, "grad_norm": 0.21695522373717682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103016893084, "tokens_per_second": 152664.82882184125, "data_wait_s": 9.547999070491642e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 703900, "loss": 0.5815778970718384, "lr": 8e-05, "grad_norm": 0.22463245838636323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103031557959, "tokens_per_second": 153466.13672966024, "data_wait_s": 9.208997653331608e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 704000, "loss": 0.5550633072853088, "lr": 8e-05, "grad_norm": 0.21852078942307152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103046306084, "tokens_per_second": 152962.1840814142, "data_wait_s": 1.9200997485313565e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 704100, "loss": 0.5729023814201355, "lr": 8e-05, "grad_norm": 0.21018539089421961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103060959449, "tokens_per_second": 152121.52362965216, "data_wait_s": 1.2620999768842012e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3db4b1a8efea6c1a6845bee5633ee71760cde6227a84e058cb30d6a36cbda802
-size 404934
+oid sha256:565ffca62455c971d3530b6087747de2a19bb35e76a396b78bea8dc3520c89fd
+size 407052