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": 77232.68716644056,
-      "required_pages_per_second_upper_estimate": 0.00534313650269815
+      "required_token_positions_per_second": 77228.23307714416,
+      "required_pages_per_second_upper_estimate": 0.005342828358465714
     },
     {
       "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": 9654.08589580507,
-      "required_pages_per_second_upper_estimate": 0.004041964164335351
+      "required_token_positions_per_second": 9653.52913464302,
+      "required_pages_per_second_upper_estimate": 0.004041731060063296
     },
     {
       "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": 19308.17179161014,
-      "required_pages_per_second_upper_estimate": 0.004021382993460654
+      "required_token_positions_per_second": 19307.05826928604,
+      "required_pages_per_second_upper_estimate": 0.0040211510761261025
     },
     {
       "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": 28962.25768741521,
-      "required_pages_per_second_upper_estimate": 0.0017362431255536369
+      "required_token_positions_per_second": 28960.587403929065,
+      "required_pages_per_second_upper_estimate": 0.0017361429946090175
     },
     {
       "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": 28962.25768741521,
-      "required_pages_per_second_upper_estimate": 0.002251688143688115
+      "required_token_positions_per_second": 28960.587403929065,
+      "required_pages_per_second_upper_estimate": 0.002251558286493868
     },
     {
       "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": 28962.25768741521,
-      "required_pages_per_second_upper_estimate": 0.37149674436468505
+      "required_token_positions_per_second": 28960.587403929065,
+      "required_pages_per_second_upper_estimate": 0.3714753197615354
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04895714574224286,
-    "producer_p95_seconds": 0.1668987630000629,
-    "gpu_update_seconds": 0.954621711414912,
-    "producer_batches_per_second": 20.426027392711056,
-    "gpu_batches_per_second": 1.0475353619580154,
-    "producer_headroom": 19.499129227037702,
+    "producer_batch_seconds": 0.06668300788287951,
+    "producer_p95_seconds": 0.16936335699938354,
+    "gpu_update_seconds": 0.9546767686158539,
+    "producer_batches_per_second": 14.996324127375551,
+    "gpu_batches_per_second": 1.0474749495055362,
+    "producer_headroom": 14.316642259038856,
     "sustainable": true,
-    "buffer_seconds": 60.14116781913946,
+    "buffer_seconds": 60.1446364227988,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330132.01783183083,
-    "upload_bytes_per_second": 5893338.212030285,
+    "generation_bytes_per_second": 328301.61756047775,
+    "upload_bytes_per_second": 5873810.8097339235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.83864594665712,
+    "estimated_full_shards_per_hour": 70.44588465796231,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 7566500,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1241090929997881,
+      "bytes": 6087377,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 488100, "loss": 0.6014329195022583, "lr": 0.0003, "grad_norm": 0.20564692473549886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71441950975, "tokens_per_second": 154479.18494264406, "data_wait_s": 9.575000149197876e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 488200, "loss": 0.5942641496658325, "lr": 0.0003, "grad_norm": 0.16873086263281123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71456706571, "tokens_per_second": 153355.76433489085, "data_wait_s": 1.0932000805041753e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488300, "loss": 0.5797544717788696, "lr": 0.0003, "grad_norm": 0.15258235446268698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71471202382, "tokens_per_second": 150730.92543615756, "data_wait_s": 1.4066000403545331e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 488400, "loss": 0.5689780116081238, "lr": 0.0003, "grad_norm": 0.17653242025923216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71485709701, "tokens_per_second": 151669.7361474753, "data_wait_s": 1.0341999768570531e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:551dabe30faedf40db90a79dee2eda9300623eb94a00935ba8316344fcdaf1c1
-size 70059
+oid sha256:949f88e712dec832cfce23856625ef7d2af75dfbae004569eeb2d6f980215384
+size 72177