Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76119.10653187743,
-      "required_pages_per_second_upper_estimate": 0.006597811232071411
+      "required_token_positions_per_second": 75992.81592299472,
+      "required_pages_per_second_upper_estimate": 0.006586864682174079
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9514.88831648468,
-      "required_pages_per_second_upper_estimate": 0.0037956312097034784
+      "required_token_positions_per_second": 9499.10199037434,
+      "required_pages_per_second_upper_estimate": 0.0037893338081914554
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19029.77663296936,
-      "required_pages_per_second_upper_estimate": 0.004402203911407407
+      "required_token_positions_per_second": 18998.20398074868,
+      "required_pages_per_second_upper_estimate": 0.004394900133975854
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28544.66494945404,
-      "required_pages_per_second_upper_estimate": 0.0017382203331390003
+      "required_token_positions_per_second": 28497.30597112302,
+      "required_pages_per_second_upper_estimate": 0.0017353364198319974
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28544.66494945404,
-      "required_pages_per_second_upper_estimate": 0.0021974473244988453
+      "required_token_positions_per_second": 28497.30597112302,
+      "required_pages_per_second_upper_estimate": 0.0021938014992488805
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28544.66494945404,
-      "required_pages_per_second_upper_estimate": 0.35442920582408133
+      "required_token_positions_per_second": 28497.30597112302,
+      "required_pages_per_second_upper_estimate": 0.35384116581351455
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10268126689058477,
-    "producer_p95_seconds": 0.3414515370022855,
-    "gpu_update_seconds": 0.9685873016536777,
-    "producer_batches_per_second": 9.738874775138695,
-    "gpu_batches_per_second": 1.0324314579518967,
-    "producer_headroom": 9.432950439594656,
+    "producer_batch_seconds": 0.08592142272652836,
+    "producer_p95_seconds": 0.34291396899789106,
+    "gpu_update_seconds": 0.9701969732863998,
+    "producer_batches_per_second": 11.63854098625451,
+    "gpu_batches_per_second": 1.03071853194166,
+    "producer_headroom": 11.291677238333834,
     "sustainable": true,
-    "buffer_seconds": 60.05241270252802,
+    "buffer_seconds": 60.15221234375679,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299325.53843147174,
-    "upload_bytes_per_second": 3788275.8141807066,
+    "generation_bytes_per_second": 297619.01196240896,
+    "upload_bytes_per_second": 3769849.5013309917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.22829260547746,
+    "estimated_full_shards_per_hour": 63.86211175111963,
     "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": 1.2876270810011192,
-      "bytes": 7494644,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8472818170012033,
       "bytes": 6269698,
@@ -562,6 +556,12 @@
       "bytes": 6506267,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.471190406999085,
+      "bytes": 6116938,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 69500, "loss": 0.6252729836851358, "lr": 0.0003, "grad_norm": 0.13367581634287012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10171816427, "tokens_per_second": 148491.94474402617, "data_wait_s": 1.2938002328155562e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69600, "loss": 0.5945924948900938, "lr": 0.0003, "grad_norm": 0.14415718360304616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10186342641, "tokens_per_second": 148896.37974293946, "data_wait_s": 1.3379001757130027e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69700, "loss": 0.6128690466284752, "lr": 0.0003, "grad_norm": 0.1299160860011532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10201051763, "tokens_per_second": 150101.95188452434, "data_wait_s": 1.0831001418409869e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 69800, "loss": 0.6174406055361032, "lr": 0.0003, "grad_norm": 0.14424695000936255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10215705395, "tokens_per_second": 143158.04213653147, "data_wait_s": 8.422000973951072e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1546671dc6ccdc274789e79eb85f0e082e00cbe937d0f3bdd2574046cc74e80c
-size 633126
+oid sha256:b82949a2eaf4af303550f68efd81c34b75c87696d0be75507dfb4c56571d33e6
+size 635244