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": 77295.40323780412,
-      "required_pages_per_second_upper_estimate": 0.005347475346036415
+      "required_token_positions_per_second": 77204.40536853319,
+      "required_pages_per_second_upper_estimate": 0.005341179902296105
     },
     {
       "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": 9661.925404725514,
-      "required_pages_per_second_upper_estimate": 0.004045246403012779
+      "required_token_positions_per_second": 9650.550671066649,
+      "required_pages_per_second_upper_estimate": 0.004040484039561262
     },
     {
       "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": 19323.85080945103,
-      "required_pages_per_second_upper_estimate": 0.004024648519393405
+      "required_token_positions_per_second": 19301.101342133297,
+      "required_pages_per_second_upper_estimate": 0.004019910405294919
     },
     {
       "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": 28985.77621417654,
-      "required_pages_per_second_upper_estimate": 0.0017376530253222675
+      "required_token_positions_per_second": 28951.652013199946,
+      "required_pages_per_second_upper_estimate": 0.0017356073315783584
     },
     {
       "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": 28985.77621417654,
-      "required_pages_per_second_upper_estimate": 0.002253516605696742
+      "required_token_positions_per_second": 28951.652013199946,
+      "required_pages_per_second_upper_estimate": 0.002250863599167309
     },
     {
       "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": 28985.77621417654,
-      "required_pages_per_second_upper_estimate": 0.3717984147737528
+      "required_token_positions_per_second": 28951.652013199946,
+      "required_pages_per_second_upper_estimate": 0.3713607061633374
     }
   ],
   "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": 0.203502714929769,
-    "producer_p95_seconds": 0.21133469000051264,
-    "gpu_update_seconds": 0.9538471488811725,
-    "producer_batches_per_second": 4.91393935626417,
-    "gpu_batches_per_second": 1.0483860031169177,
-    "producer_headroom": 4.6871470447475625,
+    "producer_batch_seconds": 0.5290212887812942,
+    "producer_p95_seconds": 0.2273545269999886,
+    "gpu_update_seconds": 0.9549714119040402,
+    "producer_batches_per_second": 1.8902830967420972,
+    "gpu_batches_per_second": 1.0471517655237248,
+    "producer_headroom": 1.805166317794142,
     "sustainable": true,
-    "buffer_seconds": 60.092370379513866,
+    "buffer_seconds": 60.163198949954534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373563.19933506014,
-    "upload_bytes_per_second": 5491201.511604676,
+    "generation_bytes_per_second": 325749.16335997835,
+    "upload_bytes_per_second": 5471085.318318932,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.15796647108891,
+    "estimated_full_shards_per_hour": 69.89818740462792,
     "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": 1.0014089890000832,
-      "bytes": 6005067,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9762548340004287,
       "bytes": 2290276,
@@ -562,6 +556,12 @@
       "bytes": 1017212,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.2100102580006933,
+      "bytes": 5596380,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 772200, "loss": 0.5448341965675354, "lr": 3e-05, "grad_norm": 0.25799923936604596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113027749587, "tokens_per_second": 152505.98524207767, "data_wait_s": 1.282700031879358e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772300, "loss": 0.5016350150108337, "lr": 3e-05, "grad_norm": 0.25983040476262986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113042303494, "tokens_per_second": 151358.02842660164, "data_wait_s": 8.4530001913663e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 772400, "loss": 0.49536532163619995, "lr": 3e-05, "grad_norm": 0.2372936969991625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113056934823, "tokens_per_second": 123204.10606463822, "data_wait_s": 1.2905999028589576e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 772500, "loss": 0.5680965185165405, "lr": 3e-05, "grad_norm": 0.25440132234263146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113071647572, "tokens_per_second": 153357.4423961609, "data_wait_s": 1.5044999599922448e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8abce7e1db3cdc2512115fa7430e68e7b08c1c55b5ffb7321be2014141c1e7c6
-size 264363
+oid sha256:f3db7ded53ffc8e0745d69f58a7447a451eebfde3713c36b150baaf8007bb4c8
+size 266481