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": 77037.86649548938,
-      "required_pages_per_second_upper_estimate": 0.0053296583566355095
+      "required_token_positions_per_second": 76549.64468686633,
+      "required_pages_per_second_upper_estimate": 0.005295882039084298
     },
     {
       "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": 9629.733311936172,
-      "required_pages_per_second_upper_estimate": 0.004031768246009222
+      "required_token_positions_per_second": 9568.705585858292,
+      "required_pages_per_second_upper_estimate": 0.004006217211504252
     },
     {
       "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": 19259.466623872344,
-      "required_pages_per_second_upper_estimate": 0.004011238991462519
+      "required_token_positions_per_second": 19137.411171716583,
+      "required_pages_per_second_upper_estimate": 0.003985818059597203
     },
     {
       "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": 28889.199935808516,
-      "required_pages_per_second_upper_estimate": 0.001731863424897543
+      "required_token_positions_per_second": 28706.116757574873,
+      "required_pages_per_second_upper_estimate": 0.0017208878679142627
     },
     {
       "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": 28889.199935808516,
-      "required_pages_per_second_upper_estimate": 0.002246008224847783
+      "required_token_positions_per_second": 28706.116757574873,
+      "required_pages_per_second_upper_estimate": 0.002231774313037912
     },
     {
       "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": 28889.199935808516,
-      "required_pages_per_second_upper_estimate": 0.37055963797037644
+      "required_token_positions_per_second": 28706.116757574873,
+      "required_pages_per_second_upper_estimate": 0.36821124353939627
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07563107155469595,
-    "producer_p95_seconds": 0.21379042199987452,
-    "gpu_update_seconds": 0.9570358494327829,
-    "producer_batches_per_second": 13.222078960983724,
-    "gpu_batches_per_second": 1.0448929374930742,
-    "producer_headroom": 12.654003569692387,
+    "producer_batch_seconds": 0.23830814819531554,
+    "producer_p95_seconds": 0.24824349000118673,
+    "gpu_update_seconds": 0.9631396762400591,
+    "producer_batches_per_second": 4.196247621295801,
+    "gpu_batches_per_second": 1.0382710054099709,
+    "producer_headroom": 4.0415725753979554,
     "sustainable": true,
-    "buffer_seconds": 60.29325851426532,
+    "buffer_seconds": 60.677799603123724,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514934.53732749156,
-    "upload_bytes_per_second": 3990288.2977678296,
+    "generation_bytes_per_second": 498435.05610526673,
+    "upload_bytes_per_second": 3960080.2558228336,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.49296464794693,
+    "estimated_full_shards_per_hour": 106.95256006592275,
     "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": 1.6416546800001015,
-      "bytes": 7689453,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.5363667999999961,
       "bytes": 6120972,
@@ -562,6 +556,12 @@
       "bytes": 5866239,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 2.011558195001271,
+      "bytes": 5842602,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 813700, "loss": 0.5054027438163757, "lr": 3e-05, "grad_norm": 0.2698917196061391, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119102279338, "tokens_per_second": 150872.49377637048, "data_wait_s": 1.0710999049479142e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813800, "loss": 0.5414049029350281, "lr": 3e-05, "grad_norm": 0.2815565741459654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119116976455, "tokens_per_second": 152972.4789200797, "data_wait_s": 1.6899000911507756e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813900, "loss": 0.529843270778656, "lr": 3e-05, "grad_norm": 0.2860852396890787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119131526765, "tokens_per_second": 150183.54668206448, "data_wait_s": 1.9241000700276345e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 814000, "loss": 0.5257862210273743, "lr": 3e-05, "grad_norm": 0.2666706961559004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119146200406, "tokens_per_second": 152149.18129974138, "data_wait_s": 8.697999874129891e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc05ab41f044bb058d1a299f3cd905bf20485600d56269de587758339552d190
-size 296133
+oid sha256:57918a817f0657d01b8ffc17d7d000c16652ecbdbde373e402c60ffd647aad15
+size 298251