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": 76942.0093304089,
-      "required_pages_per_second_upper_estimate": 0.005323026735535976
+      "required_token_positions_per_second": 76965.4638274193,
+      "required_pages_per_second_upper_estimate": 0.005324649372061087
     },
     {
       "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": 9617.751166301112,
-      "required_pages_per_second_upper_estimate": 0.004026751571847477
+      "required_token_positions_per_second": 9620.682978427412,
+      "required_pages_per_second_upper_estimate": 0.00402797906036156
     },
     {
       "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": 19235.502332602224,
-      "required_pages_per_second_upper_estimate": 0.004006247861571812
+      "required_token_positions_per_second": 19241.365956854825,
+      "required_pages_per_second_upper_estimate": 0.0040074690998694595
     },
     {
       "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": 28853.253498903337,
-      "required_pages_per_second_upper_estimate": 0.0017297084908921086
+      "required_token_positions_per_second": 28862.048935282233,
+      "required_pages_per_second_upper_estimate": 0.0017302357638731697
     },
     {
       "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": 28853.253498903337,
-      "required_pages_per_second_upper_estimate": 0.0022432135474900715
+      "required_token_positions_per_second": 28862.048935282233,
+      "required_pages_per_second_upper_estimate": 0.0022438973539814946
     },
     {
       "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": 28853.253498903337,
-      "required_pages_per_second_upper_estimate": 0.37009855567403366
+      "required_token_positions_per_second": 28862.048935282233,
+      "required_pages_per_second_upper_estimate": 0.37021137408809834
     }
   ],
   "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": 0.24915556036705766,
-    "producer_p95_seconds": 0.35227929699613014,
-    "gpu_update_seconds": 0.9582281596441404,
-    "producer_batches_per_second": 4.013556825811125,
-    "gpu_batches_per_second": 1.0435927914823255,
-    "producer_headroom": 3.845903170824172,
+    "producer_batch_seconds": 0.08947974235127276,
+    "producer_p95_seconds": 0.3297605939951609,
+    "gpu_update_seconds": 0.9579361486772989,
+    "producer_batches_per_second": 11.175713895937207,
+    "gpu_batches_per_second": 1.0439109134578355,
+    "producer_headroom": 10.70562032819346,
     "sustainable": true,
-    "buffer_seconds": 60.368374057580844,
+    "buffer_seconds": 60.34997736666983,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 570697.6114910742,
-    "upload_bytes_per_second": 3802259.5495382077,
+    "generation_bytes_per_second": 575739.4608036933,
+    "upload_bytes_per_second": 3879972.1912399833,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.22921113276086,
+    "estimated_full_shards_per_hour": 61.77014287988233,
     "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": 1.5132595840004797,
-      "bytes": 5701329,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.5889989900024375,
       "bytes": 6430899,
@@ -562,6 +556,12 @@
       "bytes": 2459570,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 2.1903198990039527,
+      "bytes": 16888519,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -354,3 +354,4 @@
 {"step": 835400, "loss": 0.5379299521446228, "lr": 3e-05, "grad_norm": 0.305677779022904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122278756085, "tokens_per_second": 152780.8836670833, "data_wait_s": 1.3860997569281608e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835500, "loss": 0.48333069682121277, "lr": 3e-05, "grad_norm": 0.29217728596827114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122293481709, "tokens_per_second": 151981.18707182675, "data_wait_s": 1.6933001461438835e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 835600, "loss": 0.5196800827980042, "lr": 3e-05, "grad_norm": 0.2856584328634872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122308166701, "tokens_per_second": 152360.93729544472, "data_wait_s": 6.574999133590609e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 835700, "loss": 0.5158669948577881, "lr": 3e-05, "grad_norm": 0.3009135458409829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122322785119, "tokens_per_second": 151778.23521017737, "data_wait_s": 1.1648997315205634e-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:bd8002bb132fab48dedc6c3d0b4bf51f0727df89e0fb0fd2bafe6b34f7cb02b8
-size 759654
+oid sha256:76a794b94226b1e9b7359e9eb3ce8371ca0357589fb5483e61affa9326602242
+size 761772