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": 76433.23624626746,
-      "required_pages_per_second_upper_estimate": 0.005287828633058873
+      "required_token_positions_per_second": 76491.07544192679,
+      "required_pages_per_second_upper_estimate": 0.0052918300828200095
     },
     {
       "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": 9554.154530783433,
-      "required_pages_per_second_upper_estimate": 0.004000124988604992
+      "required_token_positions_per_second": 9561.384430240849,
+      "required_pages_per_second_upper_estimate": 0.0040031519965303426
     },
     {
       "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": 19108.309061566866,
-      "required_pages_per_second_upper_estimate": 0.003979756857527272
+      "required_token_positions_per_second": 19122.768860481698,
+      "required_pages_per_second_upper_estimate": 0.003982768452310691
     },
     {
       "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": 28662.463592350297,
-      "required_pages_per_second_upper_estimate": 0.0017182709273135732
+      "required_token_positions_per_second": 28684.15329072255,
+      "required_pages_per_second_upper_estimate": 0.0017195711916127379
     },
     {
       "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": 28662.463592350297,
-      "required_pages_per_second_upper_estimate": 0.0022283804714516827
+      "required_token_positions_per_second": 28684.15329072255,
+      "required_pages_per_second_upper_estimate": 0.0022300667500972245
     },
     {
       "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": 28662.463592350297,
-      "required_pages_per_second_upper_estimate": 0.3676513076070124
+      "required_token_positions_per_second": 28684.15329072255,
+      "required_pages_per_second_upper_estimate": 0.367929519769148
     }
   ],
   "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": 0.08495233792962154,
-    "producer_p95_seconds": 0.2577633479995711,
-    "gpu_update_seconds": 0.964606545802258,
-    "producer_batches_per_second": 11.77130640981825,
-    "gpu_batches_per_second": 1.0366921148853552,
-    "producer_headroom": 11.35467921555476,
+    "producer_batch_seconds": 0.09748606642966706,
+    "producer_p95_seconds": 0.2203604450005514,
+    "gpu_update_seconds": 0.963877152648683,
+    "producer_batches_per_second": 10.257876193224666,
+    "gpu_batches_per_second": 1.0374766091841199,
+    "producer_headroom": 9.887332497348103,
     "sustainable": true,
-    "buffer_seconds": 60.77021238554226,
+    "buffer_seconds": 60.72426061686703,
     "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": 247953.39459260218,
-    "upload_bytes_per_second": 3851754.732492182,
+    "generation_bytes_per_second": 249036.43896383673,
+    "upload_bytes_per_second": 3909958.5875967154,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.20502641995954,
+    "estimated_full_shards_per_hour": 53.437422530043854,
     "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": 1.859183167000083,
-      "bytes": 1137645,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 4.019235966999986,
       "bytes": 13232606,
@@ -562,6 +556,12 @@
       "bytes": 6793396,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.551560215997597,
+      "bytes": 6347247,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -189,3 +189,4 @@
 {"step": 818900, "loss": 0.5745950937271118, "lr": 3e-05, "grad_norm": 0.2938284744034004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119863964638, "tokens_per_second": 98068.47779072159, "data_wait_s": 1.6866000805748627e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 819000, "loss": 0.5096691250801086, "lr": 3e-05, "grad_norm": 0.2746947830905528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119878522906, "tokens_per_second": 150749.14686199775, "data_wait_s": 1.6498997865710407e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 819100, "loss": 0.5381044149398804, "lr": 3e-05, "grad_norm": 0.27991904805261775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119893128562, "tokens_per_second": 150979.53770923213, "data_wait_s": 1.1271000403212383e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 819200, "loss": 0.5278240442276001, "lr": 3e-05, "grad_norm": 0.28942664777665683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119907919956, "tokens_per_second": 152879.20086366482, "data_wait_s": 1.1234998964937404e-05, "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 sha256:6732ca8efee36f1bc03a6be8a73cdfaa59ad28c16cf7d6ab5b8f5e786436fb48
-size 407052
+oid sha256:6a51dcfebee9b7b5179c315b4a19112db1281bf58c5ba0162ff56337c8904398
+size 409170