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": 76638.89182191226,
-      "required_pages_per_second_upper_estimate": 0.005302056362968659
+      "required_token_positions_per_second": 76381.73023113604,
+      "required_pages_per_second_upper_estimate": 0.005284265327421661
     },
     {
       "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": 9579.861477739032,
-      "required_pages_per_second_upper_estimate": 0.004010887950473205
+      "required_token_positions_per_second": 9547.716278892005,
+      "required_pages_per_second_upper_estimate": 0.003997429426984039
     },
     {
       "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": 19159.722955478064,
-      "required_pages_per_second_upper_estimate": 0.003990465015753414
+      "required_token_positions_per_second": 19095.43255778401,
+      "required_pages_per_second_upper_estimate": 0.003977075021365544
     },
     {
       "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": 28739.584433217096,
-      "required_pages_per_second_upper_estimate": 0.0017228942039668321
+      "required_token_positions_per_second": 28643.148836676017,
+      "required_pages_per_second_upper_estimate": 0.0017171130372028245
     },
     {
       "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": 28739.584433217096,
-      "required_pages_per_second_upper_estimate": 0.002234376277610347
+      "required_token_positions_per_second": 28643.148836676017,
+      "required_pages_per_second_upper_estimate": 0.0022268788341546393
     },
     {
       "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": 28739.584433217096,
-      "required_pages_per_second_upper_estimate": 0.36864053094774435
+      "required_token_positions_per_second": 28643.148836676017,
+      "required_pages_per_second_upper_estimate": 0.36740355865979163
     }
   ],
   "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": 0.1227792715000362,
-    "producer_p95_seconds": 0.23315460500089102,
-    "gpu_update_seconds": 0.9620180856910565,
-    "producer_batches_per_second": 8.144697291184897,
-    "gpu_batches_per_second": 1.0394814971505026,
-    "producer_headroom": 7.835346096598829,
+    "producer_batch_seconds": 0.38192755125803046,
+    "producer_p95_seconds": 0.24554425500173238,
+    "gpu_update_seconds": 0.9652570029101765,
+    "producer_batches_per_second": 2.618297624002515,
+    "gpu_batches_per_second": 1.0359935198450527,
+    "producer_headroom": 2.527330117271504,
     "sustainable": true,
-    "buffer_seconds": 60.60713939853656,
+    "buffer_seconds": 60.811191183341116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346306.1416333988,
-    "upload_bytes_per_second": 3796511.627884045,
+    "generation_bytes_per_second": 318270.8318310126,
+    "upload_bytes_per_second": 3798689.7506945515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.30923640014146,
+    "estimated_full_shards_per_hour": 68.29351154516014,
     "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": 1.6632540899991,
-      "bytes": 5726975,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.7110617899998033,
       "bytes": 7430334,
@@ -562,6 +556,12 @@
       "bytes": 6379384,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.7474238050017448,
+      "bytes": 6290249,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -177,3 +177,4 @@
 {"step": 817700, "loss": 0.5281907916069031, "lr": 3e-05, "grad_norm": 0.2763706457922375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119688026568, "tokens_per_second": 152241.22176873483, "data_wait_s": 7.0879978011362255e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 817800, "loss": 0.5345208048820496, "lr": 3e-05, "grad_norm": 0.28982529390367257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119702708258, "tokens_per_second": 151374.6428311945, "data_wait_s": 1.0001996997743845e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 817900, "loss": 0.5235115885734558, "lr": 3e-05, "grad_norm": 0.2744509112201299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119717615774, "tokens_per_second": 153926.00388755515, "data_wait_s": 1.0133000614587218e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 818000, "loss": 0.5703120231628418, "lr": 3e-05, "grad_norm": 0.2724048763442829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119732351605, "tokens_per_second": 151184.33777037638, "data_wait_s": 1.6276000678772107e-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 sha256:39f4efd9239888bd8f6ff3d8c49ad52f6a34e1f42d5c44c29a9a8015ae47d93b
-size 381636
+oid sha256:b90dd98a299ff628d3132d7a63e67d6a878658b844dfdbebf20f716694f3b874
+size 383754