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": 76625.08092519086,
-      "required_pages_per_second_upper_estimate": 0.005301100893087778
+      "required_token_positions_per_second": 76439.52618337888,
+      "required_pages_per_second_upper_estimate": 0.005288263785502905
     },
     {
       "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": 9578.135115648858,
-      "required_pages_per_second_upper_estimate": 0.004010165158716589
+      "required_token_positions_per_second": 9554.94077292236,
+      "required_pages_per_second_upper_estimate": 0.004000454171769958
     },
     {
       "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": 19156.270231297716,
-      "required_pages_per_second_upper_estimate": 0.003989745904361107
+      "required_token_positions_per_second": 19109.88154584472,
+      "required_pages_per_second_upper_estimate": 0.00398008436453315
     },
     {
       "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": 28734.405346946576,
-      "required_pages_per_second_upper_estimate": 0.0017225837256529214
+      "required_token_positions_per_second": 28664.82231876708,
+      "required_pages_per_second_upper_estimate": 0.0017184123293606927
     },
     {
       "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": 28734.405346946576,
-      "required_pages_per_second_upper_estimate": 0.002233973626433207
+      "required_token_positions_per_second": 28664.82231876708,
+      "required_pages_per_second_upper_estimate": 0.0022285638520556468
     },
     {
       "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": 28734.405346946576,
-      "required_pages_per_second_upper_estimate": 0.36857409918993567
+      "required_token_positions_per_second": 28664.82231876708,
+      "required_pages_per_second_upper_estimate": 0.36768156281688386
     }
   ],
   "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": 0.14150432737483243,
-    "producer_p95_seconds": 0.3290572149999207,
-    "gpu_update_seconds": 0.9621914797320829,
-    "producer_batches_per_second": 7.0669216874978575,
-    "gpu_batches_per_second": 1.0392941748750932,
-    "producer_headroom": 6.7997318356443115,
+    "producer_batch_seconds": 0.08722797266403859,
+    "producer_p95_seconds": 0.30199439200077904,
+    "gpu_update_seconds": 0.9645271717556972,
+    "producer_batches_per_second": 11.464212333026849,
+    "gpu_batches_per_second": 1.0367774276174435,
+    "producer_headroom": 11.05754429798117,
     "sustainable": true,
-    "buffer_seconds": 60.61806322312122,
+    "buffer_seconds": 60.765211820608926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 248682.09994832246,
-    "upload_bytes_per_second": 3174509.4604471615,
+    "generation_bytes_per_second": 265602.8427663836,
+    "upload_bytes_per_second": 3206098.517676635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.361389625904614,
+    "estimated_full_shards_per_hour": 56.9921871399272,
     "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": 1.4716807509976206,
-      "bytes": 1094934,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.608142429999134,
       "bytes": 7266564,
@@ -562,6 +556,12 @@
       "bytes": 8445757,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.6673135730015929,
+      "bytes": 6027005,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -317,3 +317,4 @@
 {"step": 831700, "loss": 0.5129266977310181, "lr": 3e-05, "grad_norm": 0.29703569636226895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121737547254, "tokens_per_second": 129535.01823401765, "data_wait_s": 1.7319001926807687e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 831800, "loss": 0.5605382919311523, "lr": 3e-05, "grad_norm": 0.2931005483054918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121752186886, "tokens_per_second": 151140.40237385005, "data_wait_s": 9.47800072026439e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 831900, "loss": 0.5526161789894104, "lr": 3e-05, "grad_norm": 0.2834125292201266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121766902416, "tokens_per_second": 151870.7840956441, "data_wait_s": 1.1004998668795452e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 832000, "loss": 0.5412599444389343, "lr": 3e-05, "grad_norm": 0.2851509734782647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121781713532, "tokens_per_second": 153708.75300884387, "data_wait_s": 7.790004019625485e-06, "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:f5ec2c0ae1a678ce0fdcba4e65911e580e4d245e972c0b0ef60285e006104b4e
-size 680505
+oid sha256:b173643c7ed6f99a998cf922775716f74890b2b6f9d2b1d93cb7bbea761fc440
+size 682623