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": 74812.1554004512,
-      "required_pages_per_second_upper_estimate": 0.006484528020436743
+      "required_token_positions_per_second": 73814.1191560909,
+      "required_pages_per_second_upper_estimate": 0.006398020768275329
     },
     {
       "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": 9351.5194250564,
-      "required_pages_per_second_upper_estimate": 0.003730460916330142
+      "required_token_positions_per_second": 9226.764894511363,
+      "required_pages_per_second_upper_estimate": 0.0036806944688492754
     },
     {
       "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": 18703.0388501128,
-      "required_pages_per_second_upper_estimate": 0.0043266188757321205
+      "required_token_positions_per_second": 18453.529789022727,
+      "required_pages_per_second_upper_estimate": 0.0042688993456584294
     },
     {
       "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": 28054.5582751692,
-      "required_pages_per_second_upper_estimate": 0.001708375408066046
+      "required_token_positions_per_second": 27680.294683534088,
+      "required_pages_per_second_upper_estimate": 0.0016855847189447771
     },
     {
       "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": 28054.5582751692,
-      "required_pages_per_second_upper_estimate": 0.002159717556010291
+      "required_token_positions_per_second": 27680.294683534088,
+      "required_pages_per_second_upper_estimate": 0.00213090570869829
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28054.5582751692,
-      "required_pages_per_second_upper_estimate": 0.3483437212109863
+      "required_token_positions_per_second": 27680.294683534088,
+      "required_pages_per_second_upper_estimate": 0.3436966199825433
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10509573475795264,
-    "producer_p95_seconds": 0.22055244800139917,
-    "gpu_update_seconds": 0.9855082988232596,
-    "producer_batches_per_second": 9.51513400903579,
-    "gpu_batches_per_second": 1.0147047987257378,
-    "producer_headroom": 9.377243530320204,
+    "producer_batch_seconds": 0.07827440455473322,
+    "producer_p95_seconds": 0.20095749800020712,
+    "gpu_update_seconds": 0.9988332969752197,
+    "producer_batches_per_second": 12.775568280442835,
+    "gpu_batches_per_second": 1.0011680658106947,
+    "producer_headroom": 12.760662986286755,
     "sustainable": true,
-    "buffer_seconds": 60.116006228218836,
+    "buffer_seconds": 60.9288311154884,
     "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": 311841.38333008817,
-    "upload_bytes_per_second": 5757418.377995114,
+    "generation_bytes_per_second": 360213.9645543447,
+    "upload_bytes_per_second": 5722298.038367775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.91390156676277,
+    "estimated_full_shards_per_hour": 77.29353144142871,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 5615350,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.8820894919990678,
+      "bytes": 2581402,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 365000, "loss": 0.6108045075088739, "lr": 0.0003, "grad_norm": 0.19939552588878048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53425992443, "tokens_per_second": 148894.1315876089, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365100, "loss": 0.5924281161278486, "lr": 0.0003, "grad_norm": 0.14468772941650945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53440489453, "tokens_per_second": 127387.38159713705, "data_wait_s": 1.4104000001680106e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 365200, "loss": 0.6003073621541262, "lr": 0.0003, "grad_norm": 0.18070827367960746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53454994492, "tokens_per_second": 147246.25298451452, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 365300, "loss": 0.6012932620942593, "lr": 0.0003, "grad_norm": 0.1589549066821825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53469587411, "tokens_per_second": 148940.8781855063, "data_wait_s": 9.033001333591528e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:695e1d266a654a6be87478e2743228064e7e7c500f98888e49d38ab49d5399ba
-size 217767
+oid sha256:599c746bc765c2e58ccc6c7fc32376aff19d12674b39953745b899c574622f1c
+size 219885