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": 76556.67154490015,
-      "required_pages_per_second_upper_estimate": 0.005296368173427634
+      "required_token_positions_per_second": 77463.66204649434,
+      "required_pages_per_second_upper_estimate": 0.005359115880835829
     },
     {
       "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": 9569.583943112519,
-      "required_pages_per_second_upper_estimate": 0.004006584961344412
+      "required_token_positions_per_second": 9682.957755811793,
+      "required_pages_per_second_upper_estimate": 0.00405405220920717
     },
     {
       "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": 19139.167886225037,
-      "required_pages_per_second_upper_estimate": 0.003986183936901638
+      "required_token_positions_per_second": 19365.915511623585,
+      "required_pages_per_second_upper_estimate": 0.004033409487535154
     },
     {
       "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": 28708.751829337554,
-      "required_pages_per_second_upper_estimate": 0.0017210458364429139
+      "required_token_positions_per_second": 29048.873267435378,
+      "required_pages_per_second_upper_estimate": 0.001741435597321513
     },
     {
       "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": 28708.751829337554,
-      "required_pages_per_second_upper_estimate": 0.0022319791782770045
+      "required_token_positions_per_second": 29048.873267435378,
+      "required_pages_per_second_upper_estimate": 0.00225842212405301
     },
     {
       "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": 28708.751829337554,
-      "required_pages_per_second_upper_estimate": 0.36824504341064834
+      "required_token_positions_per_second": 29048.873267435378,
+      "required_pages_per_second_upper_estimate": 0.3726077560246197
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08236491332797868,
-    "producer_p95_seconds": 0.21280864300206304,
-    "gpu_update_seconds": 0.9630512731572827,
-    "producer_batches_per_second": 12.141092117926242,
-    "gpu_batches_per_second": 1.0383663132717575,
-    "producer_headroom": 11.692494221688715,
+    "producer_batch_seconds": 0.11109052774230577,
+    "producer_p95_seconds": 0.21304043200143497,
+    "gpu_update_seconds": 0.9517752976324284,
+    "producer_batches_per_second": 9.001667561789587,
+    "gpu_batches_per_second": 1.0506681592677727,
+    "producer_headroom": 8.56756482281046,
     "sustainable": true,
-    "buffer_seconds": 60.67223020890881,
+    "buffer_seconds": 60.91361904847542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363040.57472194306,
+    "generation_bytes_per_second": 348750.083195846,
     "upload_bytes_per_second": 4719917.870342578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.90005618327827,
+    "estimated_full_shards_per_hour": 74.83364936739478,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -178,3 +178,4 @@
 {"step": 458000, "loss": 0.6119412779808044, "lr": 0.0003, "grad_norm": 0.16279702943387112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67037286837, "tokens_per_second": 151251.47856399947, "data_wait_s": 1.3622000551549718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458100, "loss": 0.6139224767684937, "lr": 0.0003, "grad_norm": 0.16996936809218305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67052086938, "tokens_per_second": 154054.12114908855, "data_wait_s": 1.1897998774657026e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458200, "loss": 0.5873919725418091, "lr": 0.0003, "grad_norm": 0.15693635068716677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67066827930, "tokens_per_second": 153708.19222548825, "data_wait_s": 1.3896002201363444e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 458300, "loss": 0.5666838884353638, "lr": 0.0003, "grad_norm": 0.1448872141843246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67081351229, "tokens_per_second": 150635.5064481098, "data_wait_s": 1.0539999493630603e-05, "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:b29446678fadbd0cb32b82a93b61ea392d8b5a50c0376ed3c78557fe130d8b11
-size 383754
+oid sha256:94dd0dfd9a444b700c70b0b518defa4b9b0959a6fabbe68196d57088a56d4f08
+size 385872