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": 76826.98596351447,
-      "required_pages_per_second_upper_estimate": 0.005315069152123238
+      "required_token_positions_per_second": 76534.31472786044,
+      "required_pages_per_second_upper_estimate": 0.005294821476949865
     },
     {
       "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": 9603.373245439308,
-      "required_pages_per_second_upper_estimate": 0.004020731836627769
+      "required_token_positions_per_second": 9566.789340982556,
+      "required_pages_per_second_upper_estimate": 0.004005414919790525
     },
     {
       "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": 19206.746490878617,
-      "required_pages_per_second_upper_estimate": 0.004000258778083328
+      "required_token_positions_per_second": 19133.57868196511,
+      "required_pages_per_second_upper_estimate": 0.003985019853051524
     },
     {
       "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": 28810.119736317924,
-      "required_pages_per_second_upper_estimate": 0.001727122687686555
+      "required_token_positions_per_second": 28700.368022947663,
+      "required_pages_per_second_upper_estimate": 0.001720543240051171
     },
     {
       "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": 28810.119736317924,
-      "required_pages_per_second_upper_estimate": 0.0022398600871744267
+      "required_token_positions_per_second": 28700.368022947663,
+      "required_pages_per_second_upper_estimate": 0.002231327374206658
     },
     {
       "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": 28810.119736317924,
-      "required_pages_per_second_upper_estimate": 0.3695452820810139
+      "required_token_positions_per_second": 28700.368022947663,
+      "required_pages_per_second_upper_estimate": 0.36813750494410874
     }
   ],
   "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": 0.12348018875758271,
-    "producer_p95_seconds": 0.26791307299936307,
-    "gpu_update_seconds": 0.9596627939434434,
-    "producer_batches_per_second": 8.098465106521727,
-    "gpu_batches_per_second": 1.042032687222147,
-    "producer_headroom": 7.7717956507781265,
+    "producer_batch_seconds": 0.07975385982027206,
+    "producer_p95_seconds": 0.270794867999939,
+    "gpu_update_seconds": 0.9633325948257446,
+    "producer_batches_per_second": 12.538578098333208,
+    "gpu_batches_per_second": 1.0380630795336974,
+    "producer_headroom": 12.07882097489258,
     "sustainable": true,
-    "buffer_seconds": 60.45875601843694,
+    "buffer_seconds": 60.68995347402191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443352.1489699318,
-    "upload_bytes_per_second": 4752941.577829811,
+    "generation_bytes_per_second": 538550.755423759,
+    "upload_bytes_per_second": 4732171.084226433,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.13305045913187,
+    "estimated_full_shards_per_hour": 115.56045529398516,
     "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": 0.9383213960009016,
-      "bytes": 933737,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 3.437801178999507,
       "bytes": 7738259,
@@ -562,6 +556,12 @@
       "bytes": 7669220,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3649325709993718,
+      "bytes": 1043801,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 463900, "loss": 0.5890750288963318, "lr": 0.0003, "grad_norm": 0.1425417499480492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67900955424, "tokens_per_second": 150741.81136325275, "data_wait_s": 1.1448002624092624e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464000, "loss": 0.5789682269096375, "lr": 0.0003, "grad_norm": 0.15111280592420623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67915593951, "tokens_per_second": 150678.78471906632, "data_wait_s": 1.577700095367618e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464100, "loss": 0.5820420384407043, "lr": 0.0003, "grad_norm": 0.13034105932482143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67930203612, "tokens_per_second": 151317.34910678276, "data_wait_s": 1.1707998055499047e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 464200, "loss": 0.5661742687225342, "lr": 0.0003, "grad_norm": 0.16984531684949464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67944768634, "tokens_per_second": 150374.72092619466, "data_wait_s": 9.531999239698052e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2acddfe949ec3be9eac0a830b7505c0da8360292d8f39e54f960d258d31a45c
-size 509499
+oid sha256:933f7c9d6ddbbfa806fdd9eb26c0d1e4f66a78839e2a64843736882cadb4c196
+size 511617