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": 75833.23133098953,
-      "required_pages_per_second_upper_estimate": 0.0052463189008167335
+      "required_token_positions_per_second": 76282.88930043497,
+      "required_pages_per_second_upper_estimate": 0.005277427282493204
     },
     {
       "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": 9479.15391637369,
-      "required_pages_per_second_upper_estimate": 0.003968723797542559
+      "required_token_positions_per_second": 9535.36116255437,
+      "required_pages_per_second_upper_estimate": 0.003992256597777639
     },
     {
       "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": 18958.30783274738,
-      "required_pages_per_second_upper_estimate": 0.003948515557362594
+      "required_token_positions_per_second": 19070.72232510874,
+      "required_pages_per_second_upper_estimate": 0.003971928531551943
     },
     {
       "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": 28437.461749121074,
-      "required_pages_per_second_upper_estimate": 0.0017047824103699034
+      "required_token_positions_per_second": 28606.083487663112,
+      "required_pages_per_second_upper_estimate": 0.0017148910261250117
     },
     {
       "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": 28437.461749121074,
-      "required_pages_per_second_upper_estimate": 0.0022108875678191267
+      "required_token_positions_per_second": 28606.083487663112,
+      "required_pages_per_second_upper_estimate": 0.0022239971663021263
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28437.461749121074,
-      "required_pages_per_second_upper_estimate": 0.3647652255502248
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28606.083487663112,
+      "required_pages_per_second_upper_estimate": 0.3669281241603252
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11432970593006075,
-    "producer_p95_seconds": 0.34066012399853207,
-    "gpu_update_seconds": 0.9722386703818433,
-    "producer_batches_per_second": 8.746633185707072,
-    "gpu_batches_per_second": 1.0285540273842981,
-    "producer_headroom": 8.50381501878955,
+    "producer_batch_seconds": 0.09182491928913805,
+    "producer_p95_seconds": 0.33435226299479837,
+    "gpu_update_seconds": 0.9665077014797813,
+    "producer_batches_per_second": 10.890289996892921,
+    "gpu_batches_per_second": 1.0346529039230004,
+    "producer_headroom": 10.525549153345231,
     "sustainable": true,
-    "buffer_seconds": 60.27879756367429,
+    "buffer_seconds": 60.88998519322622,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 565652.8129276019,
-    "upload_bytes_per_second": 5295376.215007366,
+    "generation_bytes_per_second": 568053.4180889415,
+    "upload_bytes_per_second": 5215985.901633003,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.68796296535035,
+    "estimated_full_shards_per_hour": 60.945519957548065,
     "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": 1.4537645320015145,
-      "bytes": 12446484,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 1.027312460999383,
       "bytes": 1095320,
@@ -562,6 +556,12 @@
       "bytes": 1152119,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.5365510039991932,
+      "bytes": 6323603,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 681600, "loss": 0.5432363152503967, "lr": 0.0001, "grad_norm": 0.20279183690871785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99768308648, "tokens_per_second": 151271.44525978353, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681700, "loss": 0.5722970366477966, "lr": 0.0001, "grad_norm": 0.19341553347340884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99782992451, "tokens_per_second": 152805.225464494, "data_wait_s": 1.0833995474968106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681800, "loss": 0.6396709084510803, "lr": 0.0001, "grad_norm": 0.22740485523095524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99797753860, "tokens_per_second": 154039.79191743594, "data_wait_s": 1.2878997949883342e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 681900, "loss": 0.5461499094963074, "lr": 0.0001, "grad_norm": 0.2342435415811837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99812491521, "tokens_per_second": 154134.3020605515, "data_wait_s": 9.36200376600027e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7129c85c0ec548df8db4f8d9e5dbab2010639f774c1c0a11146b855b7851591b
-size 785070
+oid sha256:6c939bf19a7f668b4b62c6afc4f791ab111d796a8ff872195429121d143172ea
+size 787188