Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76824.54561236799,
-      "required_pages_per_second_upper_estimate": 0.0053149003229685295
+      "required_token_positions_per_second": 76836.53743215596,
+      "required_pages_per_second_upper_estimate": 0.005315729944886318
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9603.068201545999,
-      "required_pages_per_second_upper_estimate": 0.004020604121119682
+      "required_token_positions_per_second": 9604.567179019496,
+      "required_pages_per_second_upper_estimate": 0.0040212317116856255
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19206.136403091998,
-      "required_pages_per_second_upper_estimate": 0.004000131712886472
+      "required_token_positions_per_second": 19209.13435803899,
+      "required_pages_per_second_upper_estimate": 0.004000756107840542
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28809.204604638,
-      "required_pages_per_second_upper_estimate": 0.0017270678269396687
+      "required_token_positions_per_second": 28813.70153705849,
+      "required_pages_per_second_upper_estimate": 0.001727337411171859
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28809.204604638,
-      "required_pages_per_second_upper_estimate": 0.002239788939711551
+      "required_token_positions_per_second": 28813.70153705849,
+      "required_pages_per_second_upper_estimate": 0.002240138556427329
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28809.204604638,
-      "required_pages_per_second_upper_estimate": 0.36953354375441566
+      "required_token_positions_per_second": 28813.70153705849,
+      "required_pages_per_second_upper_estimate": 0.36959122557507584
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1280367132187621,
-    "producer_p95_seconds": 0.3027490360000229,
-    "gpu_update_seconds": 0.9596932778751186,
-    "producer_batches_per_second": 7.810259845482063,
-    "gpu_batches_per_second": 1.0419995878413628,
-    "producer_headroom": 7.495453872167098,
+    "producer_batch_seconds": 0.09916473403151826,
+    "producer_p95_seconds": 0.2953827960009221,
+    "gpu_update_seconds": 0.9595434992772716,
+    "producer_batches_per_second": 10.084230142564216,
+    "gpu_batches_per_second": 1.042162237306803,
+    "producer_headroom": 9.676257478513406,
     "sustainable": true,
-    "buffer_seconds": 60.46067650613247,
+    "buffer_seconds": 60.451240454468106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375837.20557659597,
-    "upload_bytes_per_second": 5405638.944037515,
+    "generation_bytes_per_second": 397201.4420862074,
+    "upload_bytes_per_second": 5357477.46316246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.64591527436647,
+    "estimated_full_shards_per_hour": 85.23018309535662,
     "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": 1.121954108999489,
-      "bytes": 5958053,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 0.8286497560002317,
       "bytes": 1212322,
@@ -562,6 +556,12 @@
       "bytes": 5685962,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2225738800007093,
+      "bytes": 2613097,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 468400, "loss": 0.5596917271614075, "lr": 0.0003, "grad_norm": 0.13621850775012642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68559753609, "tokens_per_second": 150883.02814768994, "data_wait_s": 1.5162997442530468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 468500, "loss": 0.593070924282074, "lr": 0.0003, "grad_norm": 0.19331677216735058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68574480633, "tokens_per_second": 153448.52079049483, "data_wait_s": 1.1681997420964763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 468600, "loss": 0.5792874097824097, "lr": 0.0003, "grad_norm": 0.14939919336839286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68589082129, "tokens_per_second": 151018.01464970456, "data_wait_s": 1.4115998055785894e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 468700, "loss": 0.5686940550804138, "lr": 0.0003, "grad_norm": 0.16503529398332553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68603781220, "tokens_per_second": 152822.21311803354, "data_wait_s": 1.3570999726653099e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b022a523b59f82cc0b816281dfa268a5c2d1b36ee9f1f4b881ef092e07d6f00
-size 605592
+oid sha256:877e496bfb1d724da2d3e2385d7d28320404fdf4213e45957eaffce0320a431e
+size 607710