Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75115.91682564128,
-      "required_pages_per_second_upper_estimate": 0.006510857290895926
+      "required_token_positions_per_second": 75763.08410639188,
+      "required_pages_per_second_upper_estimate": 0.00656695211056091
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9389.48960320516,
-      "required_pages_per_second_upper_estimate": 0.0037456077880984365
+      "required_token_positions_per_second": 9470.385513298985,
+      "required_pages_per_second_upper_estimate": 0.0037778783761365027
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18778.97920641032,
-      "required_pages_per_second_upper_estimate": 0.004344186340656929
+      "required_token_positions_per_second": 18940.77102659797,
+      "required_pages_per_second_upper_estimate": 0.004381614030818558
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28168.46880961548,
-      "required_pages_per_second_upper_estimate": 0.0017153119619714382
+      "required_token_positions_per_second": 28411.156539896954,
+      "required_pages_per_second_upper_estimate": 0.001730090371461464
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28168.46880961548,
-      "required_pages_per_second_upper_estimate": 0.0021684867042765068
+      "required_token_positions_per_second": 28411.156539896954,
+      "required_pages_per_second_upper_estimate": 0.002187169477556209
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28168.46880961548,
-      "required_pages_per_second_upper_estimate": 0.3497581088147743
+      "required_token_positions_per_second": 28411.156539896954,
+      "required_pages_per_second_upper_estimate": 0.3527714782012858
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11535981386731464,
-    "producer_p95_seconds": 0.36589684500359,
-    "gpu_update_seconds": 0.9815229995945746,
-    "producer_batches_per_second": 8.668529936691707,
-    "gpu_batches_per_second": 1.018824826736671,
-    "producer_headroom": 8.50836150553701,
+    "producer_batch_seconds": 0.5003399182188559,
+    "producer_p95_seconds": 0.3820157649970497,
+    "gpu_update_seconds": 0.9731388428758514,
+    "producer_batches_per_second": 1.998641250851757,
+    "gpu_batches_per_second": 1.0276025947590044,
+    "producer_headroom": 1.944955434177823,
     "sustainable": true,
-    "buffer_seconds": 60.854425974863624,
+    "buffer_seconds": 60.33460825830279,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 545647.4413264928,
+    "generation_bytes_per_second": 471347.01268786454,
     "upload_bytes_per_second": 5570632.752721683,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.08323888631905,
+    "estimated_full_shards_per_hour": 101.14009652592614,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 74100, "loss": 0.6215753853321075, "lr": 0.0003, "grad_norm": 0.13346129314012523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10844556590, "tokens_per_second": 150682.29430683661, "data_wait_s": 9.87599923973903e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 74200, "loss": 0.6501174326986074, "lr": 0.0003, "grad_norm": 0.13225831017210676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10859205180, "tokens_per_second": 149691.51110297305, "data_wait_s": 1.1162999726366252e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 74300, "loss": 0.5940321460366249, "lr": 0.0003, "grad_norm": 0.1324297217732014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10873598026, "tokens_per_second": 147204.36127849793, "data_wait_s": 8.448005246464163e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 74400, "loss": 0.6435681488364935, "lr": 0.0003, "grad_norm": 0.13936324521999424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10888103339, "tokens_per_second": 148194.6666601274, "data_wait_s": 1.4944002032279968e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:632e1580e3b29264edbf3770050026bba255a97cdd40dc240347af52cf510eee
-size 731337
+oid sha256:badd7f316f5f1e5de61d8ec6e3178f1d409186f1e9aba0e181481e3c6c45a351
+size 733455