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": 76685.89059515418,
-      "required_pages_per_second_upper_estimate": 0.005305307847153714
+      "required_token_positions_per_second": 76969.9649859039,
+      "required_pages_per_second_upper_estimate": 0.005324960772649178
     },
     {
       "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": 9585.736324394273,
-      "required_pages_per_second_upper_estimate": 0.004013347626087006
+      "required_token_positions_per_second": 9621.245623237988,
+      "required_pages_per_second_upper_estimate": 0.004028214627994388
     },
     {
       "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": 19171.472648788545,
-      "required_pages_per_second_upper_estimate": 0.003992912167009737
+      "required_token_positions_per_second": 19242.491246475976,
+      "required_pages_per_second_upper_estimate": 0.004007703468021662
     },
     {
       "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": 28757.20897318282,
-      "required_pages_per_second_upper_estimate": 0.001723950768226663
+      "required_token_positions_per_second": 28863.73686971396,
+      "required_pages_per_second_upper_estimate": 0.001730336952965025
     },
     {
       "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": 28757.20897318282,
-      "required_pages_per_second_upper_estimate": 0.002235746508070524
+      "required_token_positions_per_second": 28863.73686971396,
+      "required_pages_per_second_upper_estimate": 0.0022440285834591225
     },
     {
       "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": 28757.20897318282,
-      "required_pages_per_second_upper_estimate": 0.368866599622668
+      "required_token_positions_per_second": 28863.73686971396,
+      "required_pages_per_second_upper_estimate": 0.3702330250986257
     }
   ],
   "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": 0.42754774426563813,
-    "producer_p95_seconds": 0.3034435379995557,
-    "gpu_update_seconds": 0.9614284900103762,
-    "producer_batches_per_second": 2.3389200701260013,
-    "gpu_batches_per_second": 1.0401189588101425,
-    "producer_headroom": 2.248704391276205,
+    "producer_batch_seconds": 0.10159546339843928,
+    "producer_p95_seconds": 0.3026609959997586,
+    "gpu_update_seconds": 0.957880129132219,
+    "producer_batches_per_second": 9.842959188819076,
+    "gpu_batches_per_second": 1.0439719643270386,
+    "producer_headroom": 9.428375018829177,
     "sustainable": true,
-    "buffer_seconds": 60.5699948706537,
+    "buffer_seconds": 60.346448135329794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337729.12990664155,
-    "upload_bytes_per_second": 4965227.962553354,
+    "generation_bytes_per_second": 339607.33291129686,
+    "upload_bytes_per_second": 4854957.127316951,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.46880934619365,
+    "estimated_full_shards_per_hour": 72.87182798866444,
     "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": 1.9570103289988765,
-      "bytes": 18490744,
-      "pages": 47,
-      "shards": 5
-    },
     {
       "seconds": 0.8552882639996824,
       "bytes": 1175820,
@@ -562,6 +556,12 @@
       "bytes": 5670243,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2746175409993157,
+      "bytes": 5571788,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 955100, "loss": 0.5422380566596985, "lr": 3e-05, "grad_norm": 0.3728766214485024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139800204381, "tokens_per_second": 137296.19085368796, "data_wait_s": 1.4479999663308263e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955200, "loss": 0.5147801637649536, "lr": 3e-05, "grad_norm": 0.34968766213571023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139814766801, "tokens_per_second": 146876.92102776733, "data_wait_s": 1.2184998922748491e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955300, "loss": 0.501911997795105, "lr": 3e-05, "grad_norm": 0.3489705310898255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139829334714, "tokens_per_second": 151142.41135095098, "data_wait_s": 1.4020999515196308e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 955400, "loss": 0.49536147713661194, "lr": 3e-05, "grad_norm": 0.3325124770746695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139844004871, "tokens_per_second": 151892.91272191438, "data_wait_s": 1.5466997865587473e-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 sha256:89cd4837070e79429bf3468d2fd928b403469d65462c3ab9b38e6f96b3e74af4
-size 646617
+oid sha256:eda45b45176909f304768557da4a8efbdcfdbf8c6e07a541b5a138998efd9fbd
+size 648735