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": 76248.31117994558,
-      "required_pages_per_second_upper_estimate": 0.005275035087885455
+      "required_token_positions_per_second": 76371.66331857373,
+      "required_pages_per_second_upper_estimate": 0.005283568875052145
     },
     {
       "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": 9531.038897493198,
-      "required_pages_per_second_upper_estimate": 0.003990446955655684
+      "required_token_positions_per_second": 9546.457914821716,
+      "required_pages_per_second_upper_estimate": 0.003996902576225439
     },
     {
       "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": 19062.077794986395,
-      "required_pages_per_second_upper_estimate": 0.003970128103899048
+      "required_token_positions_per_second": 19092.915829643433,
+      "required_pages_per_second_upper_estimate": 0.003976550853264446
     },
     {
       "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": 28593.11669247959,
-      "required_pages_per_second_upper_estimate": 0.0017141136865529095
+      "required_token_positions_per_second": 28639.373744465145,
+      "required_pages_per_second_upper_estimate": 0.0017168867261890147
     },
     {
       "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": 28593.11669247959,
-      "required_pages_per_second_upper_estimate": 0.0022229890550115117
+      "required_token_positions_per_second": 28639.373744465145,
+      "required_pages_per_second_upper_estimate": 0.0022265853373401195
     },
     {
       "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": 28593.11669247959,
-      "required_pages_per_second_upper_estimate": 0.3667618000343709
+      "required_token_positions_per_second": 28639.373744465145,
+      "required_pages_per_second_upper_estimate": 0.3673551358302888
     }
   ],
   "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": 0.07084041854670886,
-    "producer_p95_seconds": 0.26022898399969563,
-    "gpu_update_seconds": 0.9669460065286213,
-    "producer_batches_per_second": 14.11623506064757,
-    "gpu_batches_per_second": 1.0341839081481334,
-    "producer_headroom": 13.649637119112478,
+    "producer_batch_seconds": 0.2852437854765526,
+    "producer_p95_seconds": 0.2791147669995553,
+    "gpu_update_seconds": 0.9653842380315059,
+    "producer_batches_per_second": 3.50577313482681,
+    "gpu_batches_per_second": 1.0358569786047869,
+    "producer_headroom": 3.384418126476104,
     "sustainable": true,
-    "buffer_seconds": 60.91759841130314,
+    "buffer_seconds": 60.81920699598487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330283.34720341425,
-    "upload_bytes_per_second": 4920355.588237083,
+    "generation_bytes_per_second": 357895.7613252709,
+    "upload_bytes_per_second": 4977110.438615836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.87111770703144,
+    "estimated_full_shards_per_hour": 76.79609899347874,
     "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": 0.9580737389987917,
-      "bytes": 2350217,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.1762084339989087,
       "bytes": 10342096,
@@ -562,6 +556,12 @@
       "bytes": 5894448,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.0535939100009273,
+      "bytes": 7640568,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 420100, "loss": 0.5920270085334778, "lr": 0.0003, "grad_norm": 0.14548719408145425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61491509441, "tokens_per_second": 137382.28344275174, "data_wait_s": 1.4730001566931605e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420200, "loss": 0.5941126346588135, "lr": 0.0003, "grad_norm": 0.181814914446555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61506083731, "tokens_per_second": 150840.83369612007, "data_wait_s": 1.7055001080734655e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420300, "loss": 0.5611682534217834, "lr": 0.0003, "grad_norm": 0.14011688720046497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61520691441, "tokens_per_second": 151543.81931361408, "data_wait_s": 8.942999556893483e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 420400, "loss": 0.608125627040863, "lr": 0.0003, "grad_norm": 0.17101927890635338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61535371135, "tokens_per_second": 152043.47900095957, "data_wait_s": 1.3549997674999759e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:403c96914ec2e6f5e9dad139bb7adfb43c669fa49f0449733cb272454b9ea53c
-size 495456
+oid sha256:9e34d4fa10e1c3c592140dde15a0172ec80e0a63ab58564f6eb72eb27bf39c46
+size 497574