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": 75999.68661778489,
-      "required_pages_per_second_upper_estimate": 0.006587460216584874
+      "required_token_positions_per_second": 76168.3480918121,
+      "required_pages_per_second_upper_estimate": 0.006602079365684956
     },
     {
       "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": 9499.96082722311,
-      "required_pages_per_second_upper_estimate": 0.003789676411051185
+      "required_token_positions_per_second": 9521.043511476513,
+      "required_pages_per_second_upper_estimate": 0.003798086609014087
     },
     {
       "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": 18999.92165444622,
-      "required_pages_per_second_upper_estimate": 0.0043952974875557565
+      "required_token_positions_per_second": 19042.087022953026,
+      "required_pages_per_second_upper_estimate": 0.004405051703474665
     },
     {
       "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": 28499.88248166933,
-      "required_pages_per_second_upper_estimate": 0.0017354933158063616
+      "required_token_positions_per_second": 28563.13053442954,
+      "required_pages_per_second_upper_estimate": 0.0017393447903825713
     },
     {
       "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": 28499.88248166933,
-      "required_pages_per_second_upper_estimate": 0.0021939998461629733
+      "required_token_positions_per_second": 28563.13053442954,
+      "required_pages_per_second_upper_estimate": 0.00219886885634627
     },
     {
       "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": 28499.88248166933,
-      "required_pages_per_second_upper_estimate": 0.35387315745147363
+      "required_token_positions_per_second": 28563.13053442954,
+      "required_pages_per_second_upper_estimate": 0.3546584865891396
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.4584764918042197,
-    "producer_p95_seconds": 0.35154531699663494,
-    "gpu_update_seconds": 0.9701092633551297,
-    "producer_batches_per_second": 2.1811369129630833,
-    "gpu_batches_per_second": 1.0308117217038966,
-    "producer_headroom": 2.1159411239112984,
+    "producer_batch_seconds": 0.13186921032780674,
+    "producer_p95_seconds": 0.35438439100107644,
+    "gpu_update_seconds": 0.9679611262033601,
+    "producer_batches_per_second": 7.583271314919931,
+    "gpu_batches_per_second": 1.0330993393529204,
+    "producer_headroom": 7.340311842295532,
     "sustainable": true,
-    "buffer_seconds": 60.14677432801804,
+    "buffer_seconds": 60.013589824608324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287132.8485462612,
-    "upload_bytes_per_second": 5603544.075964946,
+    "generation_bytes_per_second": 287276.48666479444,
+    "upload_bytes_per_second": 5658346.519533827,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.612025187405365,
+    "estimated_full_shards_per_hour": 61.642846583918335,
     "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": 1.2120838849987194,
-      "bytes": 7800195,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.7076373689997126,
       "bytes": 1197583,
@@ -562,6 +556,12 @@
       "bytes": 5764775,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1778193410063977,
+      "bytes": 11708390,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 305800, "loss": 0.5321316346526146, "lr": 0.0003, "grad_norm": 0.14079146194573894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44760954425, "tokens_per_second": 149630.37791913783, "data_wait_s": 8.137001714203507e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 305900, "loss": 0.5969479642808437, "lr": 0.0003, "grad_norm": 0.15929553879314803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44775699833, "tokens_per_second": 138329.5867156669, "data_wait_s": 1.2167001841589808e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 306000, "loss": 0.6188448965549469, "lr": 0.0003, "grad_norm": 0.19175455702143535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44790385737, "tokens_per_second": 150073.86413076392, "data_wait_s": 1.3636999938171357e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 306100, "loss": 0.5415780134499073, "lr": 0.0003, "grad_norm": 0.1462976482285031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44805034880, "tokens_per_second": 150515.82325036995, "data_wait_s": 1.3244003639556468e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c182edfe6ed6bb181dd52188afb95d271a070611a688845fe3049d74311f7e72
-size 550524
+oid sha256:37689d640a0dbff3559cf581216d756e825f2f4e09b717ecff49c229f95d30fc
+size 552642