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": 76111.33882690132,
-      "required_pages_per_second_upper_estimate": 0.00659713794709115
+      "required_token_positions_per_second": 76191.06923956698,
+      "required_pages_per_second_upper_estimate": 0.006604048777186125
     },
     {
       "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": 9513.917353362665,
-      "required_pages_per_second_upper_estimate": 0.0037952438779969145
+      "required_token_positions_per_second": 9523.883654945872,
+      "required_pages_per_second_upper_estimate": 0.003799219584707943
     },
     {
       "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": 19027.83470672533,
-      "required_pages_per_second_upper_estimate": 0.004401754680947586
+      "required_token_positions_per_second": 19047.767309891744,
+      "required_pages_per_second_upper_estimate": 0.0044063657378883014
     },
     {
       "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": 28541.752060087994,
-      "required_pages_per_second_upper_estimate": 0.0017380429534593574
+      "required_token_positions_per_second": 28571.650964837616,
+      "required_pages_per_second_upper_estimate": 0.0017398636398911797
     },
     {
       "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": 28541.752060087994,
-      "required_pages_per_second_upper_estimate": 0.002197223082212053
+      "required_token_positions_per_second": 28571.650964837616,
+      "required_pages_per_second_upper_estimate": 0.0021995247826651445
     },
     {
       "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": 28541.752060087994,
-      "required_pages_per_second_upper_estimate": 0.35439303748696865
+      "required_token_positions_per_second": 28571.650964837616,
+      "required_pages_per_second_upper_estimate": 0.35476428181876174
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1313405176566107,
-    "producer_p95_seconds": 0.4362734679962159,
-    "gpu_update_seconds": 0.9686861528960657,
-    "producer_batches_per_second": 7.613796700683762,
-    "gpu_batches_per_second": 1.0323261017103587,
-    "producer_headroom": 7.375379434918112,
+    "producer_batch_seconds": 0.16226085835154436,
+    "producer_p95_seconds": 0.4056304240002646,
+    "gpu_update_seconds": 0.9676724678607362,
+    "producer_batches_per_second": 6.162915752814901,
+    "gpu_batches_per_second": 1.033407514642564,
+    "producer_headroom": 5.963683895744202,
     "sustainable": true,
-    "buffer_seconds": 60.05854147955608,
+    "buffer_seconds": 60.96336547522638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398792.76466513996,
+    "generation_bytes_per_second": 397022.74544122705,
     "upload_bytes_per_second": 5863860.475122072,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.57164387670183,
+    "estimated_full_shards_per_hour": 85.19183895518883,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 154900, "loss": 0.610124297440052, "lr": 0.0003, "grad_norm": 0.1712330672449007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22669651712, "tokens_per_second": 146885.69582243974, "data_wait_s": 1.0073003068100661e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155000, "loss": 0.602783715352416, "lr": 0.0003, "grad_norm": 0.14037592849547384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22684290888, "tokens_per_second": 150288.55313481545, "data_wait_s": 1.598800008650869e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155100, "loss": 0.5673714242875576, "lr": 0.0003, "grad_norm": 0.13810040368444318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22698961668, "tokens_per_second": 134032.77837987136, "data_wait_s": 1.4806995750404894e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 155200, "loss": 0.6262595262378454, "lr": 0.0003, "grad_norm": 0.14756991405259748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22713619019, "tokens_per_second": 148799.30874215477, "data_wait_s": 1.1732998245861381e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7c9d7c310461f311cf30c39ad27d9f271c26730e5373e266156b3fa4ca15b2f
-size 855747
+oid sha256:ea03b87e7cece5afd04c02b8e5ac26030cfe1307298e43ffb397d24b81c1d1dd
+size 857865