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": 74772.61982886393,
-      "required_pages_per_second_upper_estimate": 0.006481101177293549
+      "required_token_positions_per_second": 75308.61771470396,
+      "required_pages_per_second_upper_estimate": 0.00652756011556394
     },
     {
       "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": 9346.577478607991,
-      "required_pages_per_second_upper_estimate": 0.003728489500003188
+      "required_token_positions_per_second": 9413.577214337995,
+      "required_pages_per_second_upper_estimate": 0.0037552166963212043
     },
     {
       "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": 18693.154957215982,
-      "required_pages_per_second_upper_estimate": 0.004324332411061016
+      "required_token_positions_per_second": 18827.15442867599,
+      "required_pages_per_second_upper_estimate": 0.004355330830473137
     },
     {
       "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": 28039.73243582397,
-      "required_pages_per_second_upper_estimate": 0.0017074725922350891
+      "required_token_positions_per_second": 28240.73164301398,
+      "required_pages_per_second_upper_estimate": 0.0017197123893916211
     },
     {
       "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": 28039.73243582397,
-      "required_pages_per_second_upper_estimate": 0.0021585762218569453
+      "required_token_positions_per_second": 28240.73164301398,
+      "required_pages_per_second_upper_estimate": 0.002174049697227849
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28039.73243582397,
-      "required_pages_per_second_upper_estimate": 0.34815963390521093
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28240.73164301398,
+      "required_pages_per_second_upper_estimate": 0.35065537135743796
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.19638700953908028,
-    "producer_p95_seconds": 0.3560918139992282,
-    "gpu_update_seconds": 0.9860293803901107,
-    "producer_batches_per_second": 5.0919864931341285,
-    "gpu_batches_per_second": 1.014168563217013,
-    "producer_headroom": 5.020848286779857,
+    "producer_batch_seconds": 0.4055240378749545,
+    "producer_p95_seconds": 0.3130069740000181,
+    "gpu_update_seconds": 0.9790114629285602,
+    "producer_batches_per_second": 2.4659450651562986,
+    "gpu_batches_per_second": 1.0214384998196608,
+    "producer_headroom": 2.414188485740132,
     "sustainable": true,
-    "buffer_seconds": 60.147792203796755,
+    "buffer_seconds": 60.69871070157073,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319747.327134173,
-    "upload_bytes_per_second": 4498602.351944474,
+    "generation_bytes_per_second": 304151.6626793109,
+    "upload_bytes_per_second": 4500702.825224739,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.6103330661668,
+    "estimated_full_shards_per_hour": 65.26386652264114,
     "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": 1.1886153279992868,
-      "bytes": 11793828,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.1566395750014635,
       "bytes": 1136088,
@@ -562,6 +556,12 @@
       "bytes": 5832415,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.198589940999227,
+      "bytes": 12036533,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 185200, "loss": 0.6223879046738148, "lr": 0.0003, "grad_norm": 0.13899563103482293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27105600913, "tokens_per_second": 150016.95005225413, "data_wait_s": 7.956998160807416e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185300, "loss": 0.578156553208828, "lr": 0.0003, "grad_norm": 0.1414885166748943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27120082487, "tokens_per_second": 148110.79503216466, "data_wait_s": 9.035000402946025e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185400, "loss": 0.6030979081988335, "lr": 0.0003, "grad_norm": 0.15984949971503462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27134593787, "tokens_per_second": 148866.9656977174, "data_wait_s": 8.17600084701553e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 185500, "loss": 0.6365750879049301, "lr": 0.0003, "grad_norm": 0.13586671619894747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27149107625, "tokens_per_second": 149273.30140484785, "data_wait_s": 8.655999408802018e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94cb321086c15824235f5e738486b2523c0edef2615940d4867fa8d183fe684f
-size 648735
+oid sha256:9c863775554bf5283d6e458ef9990b5fde1fb55d1b4c56835ee87039776e28b0
+size 650853