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": 76367.13141389293,
-      "required_pages_per_second_upper_estimate": 0.005283255347370859
+      "required_token_positions_per_second": 76593.67044939786,
+      "required_pages_per_second_upper_estimate": 0.005298927843490072
     },
     {
       "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": 9545.891426736616,
-      "required_pages_per_second_upper_estimate": 0.003996665399493823
+      "required_token_positions_per_second": 9574.208806174733,
+      "required_pages_per_second_upper_estimate": 0.004008521294930438
     },
     {
       "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": 19091.782853473233,
-      "required_pages_per_second_upper_estimate": 0.003976314884206784
+      "required_token_positions_per_second": 19148.417612349465,
+      "required_pages_per_second_upper_estimate": 0.003988110410921675
     },
     {
       "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": 28637.67428020985,
-      "required_pages_per_second_upper_estimate": 0.0017167848459018415
+      "required_token_positions_per_second": 28722.6264185242,
+      "required_pages_per_second_upper_estimate": 0.0017218775968793822
     },
     {
       "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": 28637.67428020985,
-      "required_pages_per_second_upper_estimate": 0.0022264532114694238
+      "required_token_positions_per_second": 28722.6264185242,
+      "required_pages_per_second_upper_estimate": 0.002233057866558398
     },
     {
       "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": 28637.67428020985,
-      "required_pages_per_second_upper_estimate": 0.3673333369275644
+      "required_token_positions_per_second": 28722.6264185242,
+      "required_pages_per_second_upper_estimate": 0.3684230117433614
     }
   ],
   "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": 0.09358910511735985,
-    "producer_p95_seconds": 0.2728195739982766,
-    "gpu_update_seconds": 0.9654415274604277,
-    "producer_batches_per_second": 10.685004400309303,
-    "gpu_batches_per_second": 1.0357955107136085,
-    "producer_headroom": 10.315746969156006,
+    "producer_batch_seconds": 0.10785795270317067,
+    "producer_p95_seconds": 0.25219436999759637,
+    "gpu_update_seconds": 0.9625860670655405,
+    "producer_batches_per_second": 9.271453564041211,
+    "gpu_batches_per_second": 1.0388681430311124,
+    "producer_headroom": 8.924572022191217,
     "sustainable": true,
-    "buffer_seconds": 60.82281623000694,
+    "buffer_seconds": 60.642922225129055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480721.22118684347,
-    "upload_bytes_per_second": 5232229.427371365,
+    "generation_bytes_per_second": 473506.7404669678,
+    "upload_bytes_per_second": 5265081.252355272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.15158344940164,
+    "estimated_full_shards_per_hour": 101.60352383143211,
     "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": 1.246054737999657,
-      "bytes": 1155172,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.6413428979994933,
       "bytes": 11205821,
@@ -562,6 +556,12 @@
       "bytes": 11692696,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.0258839710004395,
+      "bytes": 2700953,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 742300, "loss": 0.5082201361656189, "lr": 5e-05, "grad_norm": 0.23735640135720543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108652273678, "tokens_per_second": 151401.48973396877, "data_wait_s": 1.1823998647741973e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742400, "loss": 0.48355570435523987, "lr": 5e-05, "grad_norm": 0.23324662270761445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108666971916, "tokens_per_second": 150519.64363600602, "data_wait_s": 1.3543998647946864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742500, "loss": 0.5044769048690796, "lr": 5e-05, "grad_norm": 0.22028332774413106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108681705305, "tokens_per_second": 152182.23209987266, "data_wait_s": 1.2324999261181802e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 742600, "loss": 0.5265346169471741, "lr": 5e-05, "grad_norm": 0.24166521310976496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108696442526, "tokens_per_second": 151693.52775048133, "data_wait_s": 1.4444998669205233e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa9992785a8f805ff269f1772c1719c2fb6a063f1574c845f6c5716badc2d51c
-size 479847
+oid sha256:7a0293dad75dcb1e6a52c9eabf7c9382895f603e1c07fb5d406eb731be02b405
+size 481965