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": 75319.5505841328,
-      "required_pages_per_second_upper_estimate": 0.0065285077489769285
+      "required_token_positions_per_second": 75021.1920739435,
+      "required_pages_per_second_upper_estimate": 0.006502646789496451
     },
     {
       "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": 9414.9438230166,
-      "required_pages_per_second_upper_estimate": 0.00375576185695572
+      "required_token_positions_per_second": 9377.649009242938,
+      "required_pages_per_second_upper_estimate": 0.00374088439813425
     },
     {
       "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": 18829.8876460332,
-      "required_pages_per_second_upper_estimate": 0.004355963112205745
+      "required_token_positions_per_second": 18755.298018485875,
+      "required_pages_per_second_upper_estimate": 0.004338708114605281
     },
     {
       "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": 28244.8314690498,
-      "required_pages_per_second_upper_estimate": 0.001719962047287078
+      "required_token_positions_per_second": 28132.947027728813,
+      "required_pages_per_second_upper_estimate": 0.0017131488718229266
     },
     {
       "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": 28244.8314690498,
-      "required_pages_per_second_upper_estimate": 0.002174365313185131
+      "required_token_positions_per_second": 28132.947027728813,
+      "required_pages_per_second_upper_estimate": 0.0021657521391762855
     },
     {
       "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": 28244.8314690498,
-      "required_pages_per_second_upper_estimate": 0.35070627747556776
+      "required_token_positions_per_second": 28132.947027728813,
+      "required_pages_per_second_upper_estimate": 0.34931704716749834
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14667976332037824,
-    "producer_p95_seconds": 0.411268789001042,
-    "gpu_update_seconds": 0.9788693563385642,
-    "producer_batches_per_second": 6.817573040500467,
-    "gpu_batches_per_second": 1.0215867863516277,
-    "producer_headroom": 6.673513333945841,
+    "producer_batch_seconds": 0.4199770572968191,
+    "producer_p95_seconds": 0.3770256569987396,
+    "gpu_update_seconds": 0.9827623097128491,
+    "producer_batches_per_second": 2.381082448733025,
+    "gpu_batches_per_second": 1.0175400400654229,
+    "producer_headroom": 2.3400380869335944,
     "sustainable": true,
-    "buffer_seconds": 60.68990009299098,
+    "buffer_seconds": 60.93126320219664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417111.51261794404,
-    "upload_bytes_per_second": 3779260.144207867,
+    "generation_bytes_per_second": 376542.7579028002,
+    "upload_bytes_per_second": 3812785.3773393645,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.5024207487463,
+    "estimated_full_shards_per_hour": 80.79731037915235,
     "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": 0.8663640420018055,
-      "bytes": 2740775,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9017275879996305,
       "bytes": 5703769,
@@ -562,6 +556,12 @@
       "bytes": 5561907,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.067893715997343,
+      "bytes": 7222726,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -376,3 +376,4 @@
 {"step": 352600, "loss": 0.6171035524457693, "lr": 0.0003, "grad_norm": 0.1642221486516797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51611234360, "tokens_per_second": 146443.17700148126, "data_wait_s": 1.555299968458712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352700, "loss": 0.5836805626749992, "lr": 0.0003, "grad_norm": 0.14487112585091855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51625820172, "tokens_per_second": 148252.52253081976, "data_wait_s": 1.5049998182803392e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 352800, "loss": 0.5335069727152586, "lr": 0.0003, "grad_norm": 0.13403092636276778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51640409496, "tokens_per_second": 139846.56906891658, "data_wait_s": 1.3468001270666718e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 352900, "loss": 0.6434033140540123, "lr": 0.0003, "grad_norm": 0.16423513368053935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51655041322, "tokens_per_second": 148080.79879630826, "data_wait_s": 1.726499613141641e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4ea6636653995243f3135e9ce1bf7f694e59888d4a162245bcbf3afcc931528
-size 806250
+oid sha256:8f42dac9a546e66596cc031e1ef56caa52191794f39526b4105b3768f47bf1a4
+size 808368