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": 75702.1379458047,
-      "required_pages_per_second_upper_estimate": 0.006561669451827846
+      "required_token_positions_per_second": 75224.0702947509,
+      "required_pages_per_second_upper_estimate": 0.006520231759485886
     },
     {
       "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": 9462.767243225588,
-      "required_pages_per_second_upper_estimate": 0.0037748393343009366
+      "required_token_positions_per_second": 9403.008786843862,
+      "required_pages_per_second_upper_estimate": 0.0037510007925817225
     },
     {
       "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": 18925.534486451175,
-      "required_pages_per_second_upper_estimate": 0.004378089325409546
+      "required_token_positions_per_second": 18806.017573687725,
+      "required_pages_per_second_upper_estimate": 0.004350441191067545
     },
     {
       "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": 28388.301729676765,
-      "required_pages_per_second_upper_estimate": 0.0017286986334289765
+      "required_token_positions_per_second": 28209.02636053159,
+      "required_pages_per_second_upper_estimate": 0.001717781704033205
     },
     {
       "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": 28388.301729676765,
-      "required_pages_per_second_upper_estimate": 0.002185410050999295
+      "required_token_positions_per_second": 28209.02636053159,
+      "required_pages_per_second_upper_estimate": 0.002171608936816515
     },
     {
       "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": 28388.301729676765,
-      "required_pages_per_second_upper_estimate": 0.3524876979484804
+      "required_token_positions_per_second": 28209.02636053159,
+      "required_pages_per_second_upper_estimate": 0.3502616978597612
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09331820173417782,
-    "producer_p95_seconds": 0.35844858700147597,
-    "gpu_update_seconds": 0.9739222959962109,
-    "producer_batches_per_second": 10.7160230417701,
-    "gpu_batches_per_second": 1.0267759595513877,
-    "producer_headroom": 10.436573764789037,
+    "producer_batch_seconds": 0.1390544533670095,
+    "producer_p95_seconds": 0.3524126400006935,
+    "gpu_update_seconds": 0.9801118140923665,
+    "producer_batches_per_second": 7.191427356595893,
+    "gpu_batches_per_second": 1.02029175204469,
+    "producer_headroom": 7.048402912386672,
     "sustainable": true,
-    "buffer_seconds": 60.38318235176508,
+    "buffer_seconds": 60.76693247372672,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472436.0774000067,
-    "upload_bytes_per_second": 5625806.620095818,
+    "generation_bytes_per_second": 478278.94721008313,
+    "upload_bytes_per_second": 5490552.1989092305,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.37378446102286,
+    "estimated_full_shards_per_hour": 102.62752830721732,
     "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": 1.0851846760015178,
-      "bytes": 11839223,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.710380387998157,
       "bytes": 2330946,
@@ -562,6 +556,12 @@
       "bytes": 6110509,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9653321419973508,
+      "bytes": 880569,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 74400, "loss": 0.6435681488364935, "lr": 0.0003, "grad_norm": 0.13936324521999424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10888103339, "tokens_per_second": 148194.6666601274, "data_wait_s": 1.4944002032279968e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 74500, "loss": 0.5958024971187115, "lr": 0.0003, "grad_norm": 0.13140623676506977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10902622843, "tokens_per_second": 148675.76287073368, "data_wait_s": 7.68699828768149e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 74600, "loss": 0.6414427235722542, "lr": 0.0003, "grad_norm": 0.14383066730187566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10917295654, "tokens_per_second": 149487.88130553588, "data_wait_s": 9.887000487651676e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 74700, "loss": 0.6301151644438505, "lr": 0.0003, "grad_norm": 0.1280298886391198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10931994310, "tokens_per_second": 150352.14101103327, "data_wait_s": 8.464994607493281e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12e615dfb5ef49477ea53e45838f116f8d52165b25d754e6e6a0822d227a999e
-size 737691
+oid sha256:44e25954f83c59b60b8bd4216f6e97bf9edbdd35eed1757cc8f9e00ef5b9c3e8
+size 739809