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": 75242.12785958977,
-      "required_pages_per_second_upper_estimate": 0.0065217969434396405
+      "required_token_positions_per_second": 75666.04773024593,
+      "required_pages_per_second_upper_estimate": 0.006558541243413027
     },
     {
       "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": 9405.265982448722,
-      "required_pages_per_second_upper_estimate": 0.003751901221656583
+      "required_token_positions_per_second": 9458.255966280742,
+      "required_pages_per_second_upper_estimate": 0.0037730397184780363
     },
     {
       "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": 18810.531964897444,
-      "required_pages_per_second_upper_estimate": 0.004351485516023342
+      "required_token_positions_per_second": 18916.511932561483,
+      "required_pages_per_second_upper_estimate": 0.0043760021163586895
     },
     {
       "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": 28215.797947346164,
-      "required_pages_per_second_upper_estimate": 0.0017181940581424424
+      "required_token_positions_per_second": 28374.767898842227,
+      "required_pages_per_second_upper_estimate": 0.001727874494137676
     },
     {
       "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": 28215.797947346164,
-      "required_pages_per_second_upper_estimate": 0.0021721302323144532
+      "required_token_positions_per_second": 28374.767898842227,
+      "required_pages_per_second_upper_estimate": 0.0021843681792376108
     },
     {
       "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": 28215.797947346164,
-      "required_pages_per_second_upper_estimate": 0.35034577830495506
+      "required_token_positions_per_second": 28374.767898842227,
+      "required_pages_per_second_upper_estimate": 0.3523196530643335
     }
   ],
   "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": 0.0911299797655829,
-    "producer_p95_seconds": 0.21238812400042661,
-    "gpu_update_seconds": 0.9798765943672499,
-    "producer_batches_per_second": 10.97333723295383,
-    "gpu_batches_per_second": 1.0205366734427865,
-    "producer_headroom": 10.75251631667014,
+    "producer_batch_seconds": 0.24536094420291477,
+    "producer_p95_seconds": 0.25965945899952203,
+    "gpu_update_seconds": 0.9743868248919886,
+    "producer_batches_per_second": 4.075628267769441,
+    "gpu_batches_per_second": 1.0262864546745598,
+    "producer_headroom": 3.971238487271901,
     "sustainable": true,
-    "buffer_seconds": 60.75234885076949,
+    "buffer_seconds": 60.41198314330329,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 215782.2390760599,
-    "upload_bytes_per_second": 6492781.977651867,
+    "generation_bytes_per_second": 212564.97922975628,
+    "upload_bytes_per_second": 6471484.041283021,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.30184535228107,
+    "estimated_full_shards_per_hour": 45.61149628324048,
     "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": 0.9312379190005231,
-      "bytes": 7230860,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 0.8876133860003392,
       "bytes": 811866,
@@ -562,6 +556,12 @@
       "bytes": 7568611,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.1983624430013151,
+      "bytes": 7531256,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -124,3 +124,4 @@
 {"step": 12400, "loss": 0.650935085490346, "lr": 0.0003, "grad_norm": 0.13516034694887558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1813818162, "tokens_per_second": 148622.16673962533, "data_wait_s": 8.554001396987587e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 12500, "loss": 0.6745043750852346, "lr": 0.0003, "grad_norm": 0.15191327576307453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1828371567, "tokens_per_second": 134701.158912142, "data_wait_s": 1.0660000043571927e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12600, "loss": 0.7019401229918003, "lr": 0.0003, "grad_norm": 0.1562949169520766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1842937690, "tokens_per_second": 148382.08404847415, "data_wait_s": 1.465599962102715e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 12700, "loss": 0.7312840530648828, "lr": 0.0003, "grad_norm": 0.14751725793743778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1857554894, "tokens_per_second": 148643.9782093223, "data_wait_s": 1.0300000212737359e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b1ca5ae7e2964a2af3c7abd6eec9674058c9a616028d49238ab04e608e61dfe
-size 264130
+oid sha256:03d7c52cf37bc18aa89aeb8e5ae7c6013fd92a2d043aedeb7fc2777c9117e9f2
+size 266213