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": 77008.52736352371,
-      "required_pages_per_second_upper_estimate": 0.00532762860740996
+      "required_token_positions_per_second": 77044.34298041755,
+      "required_pages_per_second_upper_estimate": 0.005330106414890357
     },
     {
       "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": 9626.065920440464,
-      "required_pages_per_second_upper_estimate": 0.004030232785773813
+      "required_token_positions_per_second": 9630.542872552194,
+      "required_pages_per_second_upper_estimate": 0.004032107192133603
     },
     {
       "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": 19252.131840880927,
-      "required_pages_per_second_upper_estimate": 0.0040097113495966424
+      "required_token_positions_per_second": 19261.08574510439,
+      "required_pages_per_second_upper_estimate": 0.0040115762117160555
     },
     {
       "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": 28878.197761321393,
-      "required_pages_per_second_upper_estimate": 0.0017312038613363878
+      "required_token_positions_per_second": 28891.62861765658,
+      "required_pages_per_second_upper_estimate": 0.001732009020665952
     },
     {
       "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": 28878.197761321393,
-      "required_pages_per_second_upper_estimate": 0.0022451528541748604
+      "required_token_positions_per_second": 28891.62861765658,
+      "required_pages_per_second_upper_estimate": 0.002246197044178828
     },
     {
       "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": 28878.197761321393,
-      "required_pages_per_second_upper_estimate": 0.37041851388927016
+      "required_token_positions_per_second": 28891.62861765658,
+      "required_pages_per_second_upper_estimate": 0.370590790493408
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09938830665615228,
-    "producer_p95_seconds": 0.30466791900107637,
-    "gpu_update_seconds": 0.9574004662102189,
-    "producer_batches_per_second": 10.061545805983389,
-    "gpu_batches_per_second": 1.0444950000477933,
-    "producer_headroom": 9.63292864544397,
+    "producer_batch_seconds": 0.08373780010154519,
+    "producer_p95_seconds": 0.3109521269980178,
+    "gpu_update_seconds": 0.9569553992918017,
+    "producer_batches_per_second": 11.942038109280915,
+    "gpu_batches_per_second": 1.0449807804418614,
+    "producer_headroom": 11.427997847224832,
     "sustainable": true,
-    "buffer_seconds": 60.31622937124379,
+    "buffer_seconds": 60.28819015538351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309302.0601897626,
-    "upload_bytes_per_second": 5342353.476098344,
+    "generation_bytes_per_second": 310097.44921872957,
+    "upload_bytes_per_second": 5510536.158994109,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.36902193326625,
+    "estimated_full_shards_per_hour": 66.5396939031736,
     "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": 3.0025762599980226,
-      "bytes": 963365,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1665040549996775,
       "bytes": 11979568,
@@ -562,6 +556,12 @@
       "bytes": 7186422,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.527223689998209,
+      "bytes": 6243254,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 635800, "loss": 0.6068922877311707, "lr": 0.0001, "grad_norm": 0.21987812565563367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93064271505, "tokens_per_second": 152040.99320863085, "data_wait_s": 1.4780998753849417e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635900, "loss": 0.5734906196594238, "lr": 0.0001, "grad_norm": 0.1751389047712576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93079051902, "tokens_per_second": 131361.38319660482, "data_wait_s": 1.3438999303616583e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 636000, "loss": 0.5701490044593811, "lr": 0.0001, "grad_norm": 0.17443959416940916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93093666340, "tokens_per_second": 151694.1968681244, "data_wait_s": 7.801998435752466e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 636100, "loss": 0.568419873714447, "lr": 0.0001, "grad_norm": 0.17273770364073185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93108323872, "tokens_per_second": 152065.60477108424, "data_wait_s": 1.0443000064697117e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6f0915ad34c361fcecd304412f319319a53be604485317bde0bb95c2735dc1e
-size 674934
+oid sha256:6aeecef33ed18b9ff7c7c2b8fd29bd3bdd9b176e0ed54ba479a1bf58771a7c72
+size 677052