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": 76190.74419368357,
-      "required_pages_per_second_upper_estimate": 0.005271052470202038
+      "required_token_positions_per_second": 76959.11837922239,
+      "required_pages_per_second_upper_estimate": 0.005324210379231356
     },
     {
       "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": 9523.843024210446,
-      "required_pages_per_second_upper_estimate": 0.0039874341937791175
+      "required_token_positions_per_second": 9619.889797402799,
+      "required_pages_per_second_upper_estimate": 0.004027646972030057
     },
     {
       "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": 19047.686048420892,
-      "required_pages_per_second_upper_estimate": 0.003967130682625333
+      "required_token_positions_per_second": 19239.779594805597,
+      "required_pages_per_second_upper_estimate": 0.0040071387024897855
     },
     {
       "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": 28571.52907263134,
-      "required_pages_per_second_upper_estimate": 0.0017128195417053938
+      "required_token_positions_per_second": 28859.669392208398,
+      "required_pages_per_second_upper_estimate": 0.0017300931138992451
     },
     {
       "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": 28571.52907263134,
-      "required_pages_per_second_upper_estimate": 0.002221310712522215
+      "required_token_positions_per_second": 28859.669392208398,
+      "required_pages_per_second_upper_estimate": 0.002243712354974008
     },
     {
       "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": 28571.52907263134,
-      "required_pages_per_second_upper_estimate": 0.36648489722593786
+      "required_token_positions_per_second": 28859.669392208398,
+      "required_pages_per_second_upper_estimate": 0.37018085186450145
     }
   ],
   "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": 0.11134113126567513,
-    "producer_p95_seconds": 0.2524390250000579,
-    "gpu_update_seconds": 0.9676765961568369,
-    "producer_batches_per_second": 8.981406858655527,
-    "gpu_batches_per_second": 1.0334031059256126,
-    "producer_headroom": 8.69109721768345,
+    "producer_batch_seconds": 0.10672622011719568,
+    "producer_p95_seconds": 0.24202216599951498,
+    "gpu_update_seconds": 0.9580151326149451,
+    "producer_batches_per_second": 9.369768730700887,
+    "gpu_batches_per_second": 1.0438248478084635,
+    "producer_headroom": 8.976380233113776,
     "sustainable": true,
-    "buffer_seconds": 60.963625557880725,
+    "buffer_seconds": 60.35495335474154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331667.56509111123,
-    "upload_bytes_per_second": 5536348.370746696,
+    "generation_bytes_per_second": 353284.15919055016,
+    "upload_bytes_per_second": 5454909.330272603,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.16813864278797,
+    "estimated_full_shards_per_hour": 75.80655652797107,
     "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": 1.3519573450003008,
-      "bytes": 11517949,
-      "pages": 22,
-      "shards": 3
-    },
     {
       "seconds": 1.325917474000562,
       "bytes": 11191562,
@@ -562,6 +556,12 @@
       "bytes": 10395259,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8835518189989671,
+      "bytes": 2669717,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -136,3 +136,4 @@
 {"step": 410700, "loss": 0.6070379018783569, "lr": 0.0003, "grad_norm": 0.16094560335790287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60115921087, "tokens_per_second": 152569.10342319703, "data_wait_s": 1.0637000741553493e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410800, "loss": 0.618433952331543, "lr": 0.0003, "grad_norm": 0.15794403351012848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60130655350, "tokens_per_second": 152182.62973155212, "data_wait_s": 1.8478998754289933e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410900, "loss": 0.5410958528518677, "lr": 0.0003, "grad_norm": 0.14483813645831878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60145173091, "tokens_per_second": 150938.0832579878, "data_wait_s": 1.2549999155453406e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 411000, "loss": 0.556839108467102, "lr": 0.0003, "grad_norm": 0.16034621578071848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60159845688, "tokens_per_second": 152360.1507626238, "data_wait_s": 9.575000149197876e-06, "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 sha256:27277227639b6ea073f1f66eb4796f9309e2e42fca8d965d2caa94a7b505afcc
-size 294798
+oid sha256:112ff8e67fa30c853d034efa6c87ce8954f2a3ad6d113f102444900b3bb09646
+size 296916