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": 76864.8744802495,
-      "required_pages_per_second_upper_estimate": 0.005317690367624439
+      "required_token_positions_per_second": 76019.68175832237,
+      "required_pages_per_second_upper_estimate": 0.005259217974004216
     },
     {
       "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": 9608.109310031188,
-      "required_pages_per_second_upper_estimate": 0.004022714727972114
+      "required_token_positions_per_second": 9502.460219790297,
+      "required_pages_per_second_upper_estimate": 0.003978481660092134
     },
     {
       "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": 19216.218620062376,
-      "required_pages_per_second_upper_estimate": 0.0040022315727954606
+      "required_token_positions_per_second": 19004.920439580594,
+      "required_pages_per_second_upper_estimate": 0.003958223734108846
     },
     {
       "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": 28824.327930093565,
-      "required_pages_per_second_upper_estimate": 0.001727974447208749
+      "required_token_positions_per_second": 28507.38065937089,
+      "required_pages_per_second_upper_estimate": 0.0017089739422793855
     },
     {
       "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": 28824.327930093565,
-      "required_pages_per_second_upper_estimate": 0.002240964711745243
+      "required_token_positions_per_second": 28507.38065937089,
+      "required_pages_per_second_upper_estimate": 0.0022163234555502656
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28824.327930093565,
-      "required_pages_per_second_upper_estimate": 0.36972752953519794
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28507.38065937089,
+      "required_pages_per_second_upper_estimate": 0.36566207025783265
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10406856290636313,
-    "producer_p95_seconds": 0.19764320899957966,
-    "gpu_update_seconds": 0.9591897534281991,
-    "producer_batches_per_second": 9.609049765583498,
-    "gpu_batches_per_second": 1.0425465831197036,
-    "producer_headroom": 9.21690207532933,
+    "producer_batch_seconds": 0.07536354776566156,
+    "producer_p95_seconds": 0.1929690879987902,
+    "gpu_update_seconds": 0.9698540995526926,
+    "producer_batches_per_second": 13.2690143928659,
+    "gpu_batches_per_second": 1.031082923154329,
+    "producer_headroom": 12.869008005944675,
     "sustainable": true,
-    "buffer_seconds": 60.428954465976545,
+    "buffer_seconds": 60.13095417226695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319736.5121122413,
-    "upload_bytes_per_second": 4056116.7174737025,
+    "generation_bytes_per_second": 325300.6547181568,
+    "upload_bytes_per_second": 4053352.2579855323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.6080124142211,
+    "estimated_full_shards_per_hour": 69.80194789083984,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 2660337,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6043403389994637,
+      "bytes": 6329322,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 808700, "loss": 0.4755537509918213, "lr": 3e-05, "grad_norm": 0.27892815376068164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118370460749, "tokens_per_second": 152427.4103789687, "data_wait_s": 1.0581999958958477e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808800, "loss": 0.5579580664634705, "lr": 3e-05, "grad_norm": 0.3085535286837493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118384829091, "tokens_per_second": 149086.49574901848, "data_wait_s": 1.5311999959521927e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808900, "loss": 0.5370275974273682, "lr": 3e-05, "grad_norm": 0.27748202659556104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118399350596, "tokens_per_second": 150115.66276547694, "data_wait_s": 1.4887000361341052e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 809000, "loss": 0.5002292990684509, "lr": 3e-05, "grad_norm": 0.2541201568625616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118413899670, "tokens_per_second": 143029.28804722716, "data_wait_s": 1.671800055191852e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c636a4242e60337f86509eae4b9984bff2a3504b931a5e37e502fdbb00682d84
-size 189450
+oid sha256:201bb61f77f223b93ba8f58e402ff9f73e55ce56db9b842ce9a3e2103302e96a
+size 191568