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": 75777.40072883423,
-      "required_pages_per_second_upper_estimate": 0.006568193039109066
+      "required_token_positions_per_second": 75177.20249018187,
+      "required_pages_per_second_upper_estimate": 0.006516169376971207
     },
     {
       "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": 9472.175091104278,
-      "required_pages_per_second_upper_estimate": 0.003778592265479607
+      "required_token_positions_per_second": 9397.150311272733,
+      "required_pages_per_second_upper_estimate": 0.0037486637590843836
     },
     {
       "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": 18944.350182208556,
-      "required_pages_per_second_upper_estimate": 0.00438244200547808
+      "required_token_positions_per_second": 18794.300622545466,
+      "required_pages_per_second_upper_estimate": 0.00434773068063208
     },
     {
       "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": 28416.525273312833,
-      "required_pages_per_second_upper_estimate": 0.0017304172991589237
+      "required_token_positions_per_second": 28191.450933818203,
+      "required_pages_per_second_upper_estimate": 0.0017167114527574974
     },
     {
       "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": 28416.525273312833,
-      "required_pages_per_second_upper_estimate": 0.002187582777516169
+      "required_token_positions_per_second": 28191.450933818203,
+      "required_pages_per_second_upper_estimate": 0.002170255931816224
     },
     {
       "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": 28416.525273312833,
-      "required_pages_per_second_upper_estimate": 0.3528381399023161
+      "required_token_positions_per_second": 28191.450933818203,
+      "required_pages_per_second_upper_estimate": 0.3500434698811503
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09679137339077215,
+    "producer_batch_seconds": 0.14691339625022692,
     "producer_p95_seconds": 0.39160335599444807,
-    "gpu_update_seconds": 0.972954987778376,
-    "producer_batches_per_second": 10.331499233539521,
-    "gpu_batches_per_second": 1.0277967763785023,
-    "producer_headroom": 10.052083710500744,
+    "gpu_update_seconds": 0.9807228462595274,
+    "producer_batches_per_second": 6.8067312139239675,
+    "gpu_batches_per_second": 1.0196560667613643,
+    "producer_headroom": 6.675516809843082,
     "sustainable": true,
-    "buffer_seconds": 60.32320924225931,
+    "buffer_seconds": 60.8048164680907,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 541944.8395234937,
-    "upload_bytes_per_second": 5460739.605808865,
+    "generation_bytes_per_second": 539487.7810392154,
+    "upload_bytes_per_second": 5467670.293779355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.2887467315541,
+    "estimated_full_shards_per_hour": 115.76151917822214,
     "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": 1.5226552149979398,
-      "bytes": 6060968,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.020731942000566,
       "bytes": 2578449,
@@ -562,6 +556,12 @@
       "bytes": 10631332,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8141304599994328,
+      "bytes": 2723440,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 73800, "loss": 0.658874835819006, "lr": 0.0003, "grad_norm": 0.15744462732086859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10800361514, "tokens_per_second": 150753.17500461583, "data_wait_s": 1.0556999768596143e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 73900, "loss": 0.6067108102142811, "lr": 0.0003, "grad_norm": 0.12604508464039318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10815138511, "tokens_per_second": 151301.4488370125, "data_wait_s": 1.397199957864359e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 74000, "loss": 0.665970329195261, "lr": 0.0003, "grad_norm": 0.1809262095554893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10829789275, "tokens_per_second": 149308.22611038256, "data_wait_s": 8.237999281845987e-06, "gpu_peak_gb": 85.46427917480469}
+{"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}

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:f2df6f1a37e7d92e1cd46e5bc0185765f3d6ebee8a8a3bf02cd29000ed51925a
-size 724983
+oid sha256:dae076364c44ca0cf1ba6cdf5997d0dd68dae453197c8c76b2c5391cef4851c2
+size 727101