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": 75251.37902276638,
-      "required_pages_per_second_upper_estimate": 0.00652259881081693
+      "required_token_positions_per_second": 75592.67954271586,
+      "required_pages_per_second_upper_estimate": 0.0065521818748678915
     },
     {
       "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": 9406.422377845798,
-      "required_pages_per_second_upper_estimate": 0.0037523625250701284
+      "required_token_positions_per_second": 9449.084942839483,
+      "required_pages_per_second_upper_estimate": 0.0037693812601082986
     },
     {
       "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": 18812.844755691596,
-      "required_pages_per_second_upper_estimate": 0.004352020539469841
+      "required_token_positions_per_second": 18898.169885678966,
+      "required_pages_per_second_upper_estimate": 0.00437175900662142
     },
     {
       "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": 28219.267133537396,
-      "required_pages_per_second_upper_estimate": 0.001718405313380075
+      "required_token_positions_per_second": 28347.25482851845,
+      "required_pages_per_second_upper_estimate": 0.0017261990925048819
     },
     {
       "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": 28219.267133537396,
-      "required_pages_per_second_upper_estimate": 0.0021723972998707766
+      "required_token_positions_per_second": 28347.25482851845,
+      "required_pages_per_second_upper_estimate": 0.0021822501469230334
     },
     {
       "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": 28219.267133537396,
-      "required_pages_per_second_upper_estimate": 0.35038885398683084
+      "required_token_positions_per_second": 28347.25482851845,
+      "required_pages_per_second_upper_estimate": 0.3519780328112352
     }
   ],
   "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": 0.09110264685165248,
-    "producer_p95_seconds": 0.35452154200174846,
-    "gpu_update_seconds": 0.9797561314815838,
-    "producer_batches_per_second": 10.976629489463196,
-    "gpu_batches_per_second": 1.020662150373893,
-    "producer_headroom": 10.754420045303133,
+    "producer_batch_seconds": 0.12237359335921383,
+    "producer_p95_seconds": 0.3655357300012838,
+    "gpu_update_seconds": 0.9753325380976583,
+    "producer_batches_per_second": 8.171697606889856,
+    "gpu_batches_per_second": 1.0252913349435204,
+    "producer_headroom": 7.970122567494443,
     "sustainable": true,
-    "buffer_seconds": 60.7448801518582,
+    "buffer_seconds": 60.47061736205482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258719.17182395628,
-    "upload_bytes_per_second": 5184088.479685348,
+    "generation_bytes_per_second": 260563.780839372,
+    "upload_bytes_per_second": 5194207.33585074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.51511159934059,
+    "estimated_full_shards_per_hour": 55.9109217537486,
     "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": 1.8479619009995076,
-      "bytes": 5907251,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.345177949002391,
       "bytes": 6941210,
@@ -562,6 +556,12 @@
       "bytes": 7427910,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.7726261329953559,
+      "bytes": 1134173,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -334,3 +334,4 @@
 {"step": 270500, "loss": 0.5988229159265757, "lr": 0.0003, "grad_norm": 0.1474321069976209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39593205699, "tokens_per_second": 150691.4202247918, "data_wait_s": 1.7910002497956157e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270600, "loss": 0.58234653621912, "lr": 0.0003, "grad_norm": 0.15620423272467027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39607869036, "tokens_per_second": 149560.0738369881, "data_wait_s": 1.1085001460742205e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270700, "loss": 0.6026381179690361, "lr": 0.0003, "grad_norm": 0.15153809280691205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39622440658, "tokens_per_second": 148662.09614779672, "data_wait_s": 1.6328995116055012e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 270800, "loss": 0.6145585998892784, "lr": 0.0003, "grad_norm": 0.1448784433924008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39636939876, "tokens_per_second": 147629.57016237968, "data_wait_s": 1.2777003576047719e-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 sha256:3104c8aa424358f21b949de64a40fb7a7b15fdc14a9aad23a8f0eb8d6d845b3e
-size 717294
+oid sha256:277aea7f0275222846c73616d4b60627029f730b3ae10b4e700f0395c56a1caa
+size 719412