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": 75032.6089949876,
-      "required_pages_per_second_upper_estimate": 0.006503636379276632
+      "required_token_positions_per_second": 74662.74045688349,
+      "required_pages_per_second_upper_estimate": 0.006471577111817007
     },
     {
       "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": 9379.07612437345,
-      "required_pages_per_second_upper_estimate": 0.003741453695697084
+      "required_token_positions_per_second": 9332.842557110436,
+      "required_pages_per_second_upper_estimate": 0.0037230104344624366
     },
     {
       "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": 18758.1522487469,
-      "required_pages_per_second_upper_estimate": 0.004339368390543432
+      "required_token_positions_per_second": 18665.685114220872,
+      "required_pages_per_second_upper_estimate": 0.004317977746336822
     },
     {
       "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": 28137.22837312035,
-      "required_pages_per_second_upper_estimate": 0.001713409583294788
+      "required_token_positions_per_second": 27998.52767133131,
+      "required_pages_per_second_upper_estimate": 0.0017049634382621777
     },
     {
       "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": 28137.22837312035,
-      "required_pages_per_second_upper_estimate": 0.0021660817289960482
+      "required_token_positions_per_second": 27998.52767133131,
+      "required_pages_per_second_upper_estimate": 0.002155404165024212
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28137.22837312035,
-      "required_pages_per_second_upper_estimate": 0.3493702071485199
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27998.52767133131,
+      "required_pages_per_second_upper_estimate": 0.3476480086336878
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12178517240624842,
-    "producer_p95_seconds": 0.40774797000631224,
-    "gpu_update_seconds": 0.9826127731334152,
-    "producer_batches_per_second": 8.211180230251848,
-    "gpu_batches_per_second": 1.0176948919676052,
-    "producer_headroom": 8.068410576746045,
+    "producer_batch_seconds": 0.13071090790640483,
+    "producer_p95_seconds": 0.4467034029948991,
+    "gpu_update_seconds": 0.9874804962801588,
+    "producer_batches_per_second": 7.6504709210347395,
+    "gpu_batches_per_second": 1.012678228853129,
+    "producer_headroom": 7.554690821880309,
     "sustainable": true,
-    "buffer_seconds": 60.921991934271745,
+    "buffer_seconds": 60.236310273089686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351618.55696328054,
-    "upload_bytes_per_second": 3720493.4327258384,
+    "generation_bytes_per_second": 352938.3446217756,
+    "upload_bytes_per_second": 3665479.853821214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.44915706323445,
+    "estimated_full_shards_per_hour": 75.7323527716632,
     "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": 1.0585707080026623,
-      "bytes": 7490877,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 33.798858237001696,
       "bytes": 12238534,
@@ -562,6 +556,12 @@
       "bytes": 16710893,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0522764270062908,
+      "bytes": 1123327,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -412,3 +412,4 @@
 {"step": 41200, "loss": 0.6779820397496223, "lr": 0.0003, "grad_norm": 0.12701036529413634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6028842649, "tokens_per_second": 148187.00278750827, "data_wait_s": 1.006700040306896e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41300, "loss": 0.6393550038337708, "lr": 0.0003, "grad_norm": 0.1304140443170049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6043338175, "tokens_per_second": 146804.45269256743, "data_wait_s": 1.0766001651063561e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41400, "loss": 0.6243662759661674, "lr": 0.0003, "grad_norm": 0.1266594176148852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6057938636, "tokens_per_second": 147845.282010544, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 41500, "loss": 0.6293739005923271, "lr": 0.0003, "grad_norm": 0.13255592951291514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6072666804, "tokens_per_second": 148697.67904327752, "data_wait_s": 1.361100294161588e-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 sha256:823736ea1d1c059304fc6779d24e267c0d2cad249b6b7b61e9d83b3a4092147f
-size 877507
+oid sha256:c6c592769cb04e9535b0f94aaa09398025e61e61b6adf06d9d2b77dcf4229096
+size 879625