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": 74449.93344156019,
-      "required_pages_per_second_upper_estimate": 0.006453131539083233
+      "required_token_positions_per_second": 74895.39808662326,
+      "required_pages_per_second_upper_estimate": 0.006491743285497478
     },
     {
       "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": 9306.241680195024,
-      "required_pages_per_second_upper_estimate": 0.0037123989469423266
+      "required_token_positions_per_second": 9361.924760827907,
+      "required_pages_per_second_upper_estimate": 0.00373461176034303
     },
     {
       "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": 18612.483360390048,
-      "required_pages_per_second_upper_estimate": 0.004305670456906949
+      "required_token_positions_per_second": 18723.849521655815,
+      "required_pages_per_second_upper_estimate": 0.004331433058338287
     },
     {
       "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": 27918.725040585072,
-      "required_pages_per_second_upper_estimate": 0.001700103876741777
+      "required_token_positions_per_second": 28085.774282483722,
+      "required_pages_per_second_upper_estimate": 0.0017102762991337674
     },
     {
       "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": 27918.725040585072,
-      "required_pages_per_second_upper_estimate": 0.002149260737601546
+      "required_token_positions_per_second": 28085.774282483722,
+      "required_pages_per_second_upper_estimate": 0.002162120650664802
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27918.725040585072,
-      "required_pages_per_second_upper_estimate": 0.34665712704204366
+      "required_token_positions_per_second": 28085.774282483722,
+      "required_pages_per_second_upper_estimate": 0.3487313195485767
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.050797724234385555,
-    "producer_p95_seconds": 0.15812268199988466,
-    "gpu_update_seconds": 0.990303101585351,
-    "producer_batches_per_second": 19.68592127052591,
-    "gpu_batches_per_second": 1.0097918489794948,
-    "producer_headroom": 19.49502889176684,
+    "producer_batch_seconds": 0.06896572457030814,
+    "producer_p95_seconds": 0.16405840299967167,
+    "gpu_update_seconds": 0.9844129530458859,
+    "producer_batches_per_second": 14.49995640922376,
+    "gpu_batches_per_second": 1.0158338499162227,
+    "producer_headroom": 14.273944907840582,
     "sustainable": true,
-    "buffer_seconds": 60.408489196706405,
+    "buffer_seconds": 60.049190135799044,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 174,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 613982.4749310212,
-    "upload_bytes_per_second": 4408656.15435845,
+    "generation_bytes_per_second": 633240.9504920006,
+    "upload_bytes_per_second": 4460939.630275791,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.87317316984166,
+    "estimated_full_shards_per_hour": 67.93938344035155,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 1180754,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.0497820410000713,
+      "bytes": 5934890,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 239100, "loss": 0.5797880366444588, "lr": 0.0003, "grad_norm": 0.15566681785785072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34996790354, "tokens_per_second": 147708.91423019825, "data_wait_s": 1.8267000086780172e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239200, "loss": 0.5632160399109125, "lr": 0.0003, "grad_norm": 0.14221490164697906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35011456946, "tokens_per_second": 149026.89443300694, "data_wait_s": 1.245299972651992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239300, "loss": 0.5887480089440942, "lr": 0.0003, "grad_norm": 0.1651936403064816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35026179444, "tokens_per_second": 149111.379039463, "data_wait_s": 1.3530999694921775e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 239400, "loss": 0.6398460641503334, "lr": 0.0003, "grad_norm": 0.17507099564559192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35040837015, "tokens_per_second": 148336.3790418062, "data_wait_s": 1.5294999684556387e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c69e74ea68ae6c631cfcda9b2074fef0644a051b9af54850af4473a04d37585
-size 46761
+oid sha256:e873907747675db82a77ad4c43d592d6fd0fa49b84080ab96bcb897ed652134a
+size 48879