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": 77359.74859878847,
-      "required_pages_per_second_upper_estimate": 0.005351926907411119
+      "required_token_positions_per_second": 77564.56429292778,
+      "required_pages_per_second_upper_estimate": 0.005366096532369562
     },
     {
       "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": 9669.968574848559,
-      "required_pages_per_second_upper_estimate": 0.004048613910382806
+      "required_token_positions_per_second": 9695.570536615973,
+      "required_pages_per_second_upper_estimate": 0.004059332917145066
     },
     {
       "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": 19339.937149697118,
-      "required_pages_per_second_upper_estimate": 0.004027998879841345
+      "required_token_positions_per_second": 19391.141073231945,
+      "required_pages_per_second_upper_estimate": 0.00403866330677538
     },
     {
       "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": 29009.905724545675,
-      "required_pages_per_second_upper_estimate": 0.001739099552625268
+      "required_token_positions_per_second": 29086.711609847916,
+      "required_pages_per_second_upper_estimate": 0.001743703948173341
     },
     {
       "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": 29009.905724545675,
-      "required_pages_per_second_upper_estimate": 0.002255392568993443
+      "required_token_positions_per_second": 29086.711609847916,
+      "required_pages_per_second_upper_estimate": 0.002261363888742306
     },
     {
       "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": 29009.905724545675,
-      "required_pages_per_second_upper_estimate": 0.37210792222451833
+      "required_token_positions_per_second": 29086.711609847916,
+      "required_pages_per_second_upper_estimate": 0.37309310565344106
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05593152137511481,
-    "producer_p95_seconds": 0.19013399399955233,
-    "gpu_update_seconds": 0.953053769375288,
-    "producer_batches_per_second": 17.879005888170287,
-    "gpu_batches_per_second": 1.0492587429306162,
-    "producer_headroom": 17.03965395440366,
+    "producer_batch_seconds": 0.10413156476565177,
+    "producer_p95_seconds": 0.19439534099910816,
+    "gpu_update_seconds": 0.95053715149564,
+    "producer_batches_per_second": 9.603236081686676,
+    "gpu_batches_per_second": 1.0520367335737817,
+    "producer_headroom": 9.128232670226604,
     "sustainable": true,
-    "buffer_seconds": 60.04238747064314,
+    "buffer_seconds": 60.83437769572096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283746.3280339895,
-    "upload_bytes_per_second": 5090470.609029983,
+    "generation_bytes_per_second": 297466.94827246736,
+    "upload_bytes_per_second": 5120444.1435281765,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.88535671963467,
+    "estimated_full_shards_per_hour": 63.82948242311969,
     "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": 1.5373412039998584,
-      "bytes": 6259058,
-      "pages": 42,
-      "shards": 2
-    },
     {
       "seconds": 1.4518450940004186,
       "bytes": 6815038,
@@ -562,6 +556,12 @@
       "bytes": 6113163,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3073732009997912,
+      "bytes": 7581885,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -114,3 +114,4 @@
 {"step": 656400, "loss": 0.5306210517883301, "lr": 0.0001, "grad_norm": 0.19070432796215145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96079568729, "tokens_per_second": 154493.6201685895, "data_wait_s": 1.0431000191601925e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656500, "loss": 0.5420908331871033, "lr": 0.0001, "grad_norm": 0.19473386012472932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96094250884, "tokens_per_second": 153851.7754987289, "data_wait_s": 1.003199940896593e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656600, "loss": 0.6092599630355835, "lr": 0.0001, "grad_norm": 0.1899760340959894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96108834698, "tokens_per_second": 152279.08574412647, "data_wait_s": 1.2243999663041905e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 656700, "loss": 0.540803074836731, "lr": 0.0001, "grad_norm": 0.18111740985091815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96123322377, "tokens_per_second": 138456.61213399825, "data_wait_s": 8.408998837694526e-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:a3141e71048c65137e462f95f3c422e0b2aa74578b3c780578f96ecc1c8c0c1e
-size 247419
+oid sha256:468cd370f6388effaf2db0be267c0917433bd888b1c1ac7130c325852b969248
+size 249537