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": 76828.87562130124,
-      "required_pages_per_second_upper_estimate": 0.0053151998830335365
+      "required_token_positions_per_second": 77383.01933907218,
+      "required_pages_per_second_upper_estimate": 0.005353536831219504
     },
     {
       "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": 9603.609452662655,
-      "required_pages_per_second_upper_estimate": 0.004020830731659617
+      "required_token_positions_per_second": 9672.877417384023,
+      "required_pages_per_second_upper_estimate": 0.004049831782008866
     },
     {
       "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": 19207.21890532531,
-      "required_pages_per_second_upper_estimate": 0.004000357169554167
+      "required_token_positions_per_second": 19345.754834768046,
+      "required_pages_per_second_upper_estimate": 0.004029210550218947
     },
     {
       "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": 28810.828357987968,
-      "required_pages_per_second_upper_estimate": 0.0017271651684736716
+      "required_token_positions_per_second": 29018.63225215207,
+      "required_pages_per_second_upper_estimate": 0.0017396226946306353
     },
     {
       "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": 28810.828357987968,
-      "required_pages_per_second_upper_estimate": 0.0022399151793923754
+      "required_token_positions_per_second": 29018.63225215207,
+      "required_pages_per_second_upper_estimate": 0.002256071018130902
     },
     {
       "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": 28810.828357987968,
-      "required_pages_per_second_upper_estimate": 0.36955437151893855
+      "required_token_positions_per_second": 29018.63225215207,
+      "required_pages_per_second_upper_estimate": 0.37221985675083785
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.059246873984278636,
+    "producer_batch_seconds": 0.07985731862500245,
     "producer_p95_seconds": 0.1925121410004067,
-    "gpu_update_seconds": 0.9596391903926093,
-    "producer_batches_per_second": 16.87852763785231,
-    "gpu_batches_per_second": 1.0420583173462084,
-    "producer_headroom": 16.19729659740787,
+    "gpu_update_seconds": 0.9527671655837459,
+    "producer_batches_per_second": 12.522333797555168,
+    "gpu_batches_per_second": 1.0495743725460096,
+    "producer_headroom": 11.930868478790183,
     "sustainable": true,
-    "buffer_seconds": 60.457268994734385,
+    "buffer_seconds": 60.02433143177599,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294502.53526424634,
+    "generation_bytes_per_second": 342377.0960118865,
     "upload_bytes_per_second": 5447569.995296784,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.19338839955847,
+    "estimated_full_shards_per_hour": 73.46615467326589,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 616700, "loss": 0.5233221650123596, "lr": 0.0001, "grad_norm": 0.14183613903834452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90266963749, "tokens_per_second": 151179.00985066686, "data_wait_s": 1.564699959999416e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616800, "loss": 0.5578149557113647, "lr": 0.0001, "grad_norm": 0.15761958763888476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90281471370, "tokens_per_second": 151104.57455760703, "data_wait_s": 1.6025000149966218e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616900, "loss": 0.524788498878479, "lr": 0.0001, "grad_norm": 0.1643309147252453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90296100518, "tokens_per_second": 152226.6330800492, "data_wait_s": 1.485800021328032e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 617000, "loss": 0.5374523401260376, "lr": 0.0001, "grad_norm": 0.15429343033973236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90310856570, "tokens_per_second": 130200.13436249521, "data_wait_s": 1.4109999028733e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d963d9a436baaabc3905e89c2dcd854711a56b54af0af9593f275972de0cab6
-size 267264
+oid sha256:ce6b954016c36c1f42548a0d44c414f8337270fa5d53f88748777e0de45940b2
+size 269382