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": 77231.78101782064,
-      "required_pages_per_second_upper_estimate": 0.005343073813234069
+      "required_token_positions_per_second": 77101.55282287365,
+      "required_pages_per_second_upper_estimate": 0.005334064324536592
     },
     {
       "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": 9653.97262722758,
-      "required_pages_per_second_upper_estimate": 0.004041916741147273
+      "required_token_positions_per_second": 9637.694102859206,
+      "required_pages_per_second_upper_estimate": 0.004035101262928479
     },
     {
       "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": 19307.94525445516,
-      "required_pages_per_second_upper_estimate": 0.004021335811745458
+      "required_token_positions_per_second": 19275.388205718413,
+      "required_pages_per_second_upper_estimate": 0.004014555037080706
     },
     {
       "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": 28961.91788168274,
-      "required_pages_per_second_upper_estimate": 0.0017362227547188286
+      "required_token_positions_per_second": 28913.082308577617,
+      "required_pages_per_second_upper_estimate": 0.001733295136678779
     },
     {
       "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": 28961.91788168274,
-      "required_pages_per_second_upper_estimate": 0.0022516617252870633
+      "required_token_positions_per_second": 28913.082308577617,
+      "required_pages_per_second_upper_estimate": 0.0022478649742831247
     },
     {
       "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": 28961.91788168274,
-      "required_pages_per_second_upper_estimate": 0.37149238570160387
+      "required_token_positions_per_second": 28913.082308577617,
+      "required_pages_per_second_upper_estimate": 0.3708659754053644
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1161416496796619,
-    "producer_p95_seconds": 0.30389670799922897,
-    "gpu_update_seconds": 0.9546329118447732,
-    "producer_batches_per_second": 8.610175615364232,
-    "gpu_batches_per_second": 1.0475230715307704,
-    "producer_headroom": 8.219557019190018,
+    "producer_batch_seconds": 0.08090730296902393,
+    "producer_p95_seconds": 0.28614872899925103,
+    "gpu_update_seconds": 0.9562453323005861,
+    "producer_batches_per_second": 12.359823690858398,
+    "gpu_batches_per_second": 1.0457567385914937,
+    "producer_headroom": 11.819023712441545,
     "sustainable": true,
-    "buffer_seconds": 60.14187344622071,
+    "buffer_seconds": 60.243455934936925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382141.6720830961,
-    "upload_bytes_per_second": 5129628.357117846,
+    "generation_bytes_per_second": 381952.8613487281,
+    "upload_bytes_per_second": 5049816.024090298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.9987070261923,
+    "estimated_full_shards_per_hour": 81.95819263788589,
     "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": 1.3469298159980099,
-      "bytes": 8271893,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2550481799989939,
       "bytes": 6009424,
@@ -562,6 +556,12 @@
       "bytes": 7894203,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2813600879999285,
+      "bytes": 1187408,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 952800, "loss": 0.5311126112937927, "lr": 3e-05, "grad_norm": 0.34549015143665845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139463934809, "tokens_per_second": 152711.52280609563, "data_wait_s": 1.7214002582477406e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952900, "loss": 0.5499350428581238, "lr": 3e-05, "grad_norm": 0.354671723508823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139478697578, "tokens_per_second": 153814.60586857508, "data_wait_s": 1.2526001228252426e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 953000, "loss": 0.4624035656452179, "lr": 3e-05, "grad_norm": 0.35700770549621713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139493468077, "tokens_per_second": 153853.7115908976, "data_wait_s": 1.03069978649728e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 953100, "loss": 0.5442652702331543, "lr": 3e-05, "grad_norm": 0.333775333286914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139508205139, "tokens_per_second": 153124.25215727597, "data_wait_s": 1.4394998288480565e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ddddc944060fcbf34201b68281208898926bdc5a2fbc0edf3d060745225fea79
-size 597120
+oid sha256:074dbef6ce67829028e97bef9b7f08581d4d5c0471d9060d8c65afd8bb84baa3
+size 599238