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": 74533.35706315227,
-      "required_pages_per_second_upper_estimate": 0.006460362487167599
+      "required_token_positions_per_second": 74881.81764898654,
+      "required_pages_per_second_upper_estimate": 0.006490566167582432
     },
     {
       "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": 9316.669632894034,
-      "required_pages_per_second_upper_estimate": 0.0037165588131857484
+      "required_token_positions_per_second": 9360.227206123318,
+      "required_pages_per_second_upper_estimate": 0.0037339345803906647
     },
     {
       "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": 18633.33926578807,
-      "required_pages_per_second_upper_estimate": 0.004310495103569382
+      "required_token_positions_per_second": 18720.454412246636,
+      "required_pages_per_second_upper_estimate": 0.004330647659528359
     },
     {
       "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": 27950.008898682103,
-      "required_pages_per_second_upper_estimate": 0.0017020088995661677
+      "required_token_positions_per_second": 28080.681618369956,
+      "required_pages_per_second_upper_estimate": 0.0017099661826083827
     },
     {
       "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": 27950.008898682103,
-      "required_pages_per_second_upper_estimate": 0.002151669055597117
+      "required_token_positions_per_second": 28080.681618369956,
+      "required_pages_per_second_upper_estimate": 0.0021617286032839275
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27950.008898682103,
-      "required_pages_per_second_upper_estimate": 0.34704556785927093
+      "required_token_positions_per_second": 28080.681618369956,
+      "required_pages_per_second_upper_estimate": 0.3486680857043341
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10526933074223166,
+    "producer_batch_seconds": 0.3073267972341398,
     "producer_p95_seconds": 0.3487805330005358,
-    "gpu_update_seconds": 0.9891946761170318,
-    "producer_batches_per_second": 9.499442933181134,
-    "gpu_batches_per_second": 1.0109233542636755,
-    "producer_headroom": 9.396798375580339,
+    "gpu_update_seconds": 0.9845914844856579,
+    "producer_batches_per_second": 3.253865295834065,
+    "gpu_batches_per_second": 1.0156496534422004,
+    "producer_headroom": 3.2037280619416264,
     "sustainable": true,
-    "buffer_seconds": 60.34087524313894,
+    "buffer_seconds": 60.060080553625134,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405451.0137505994,
-    "upload_bytes_per_second": 4115271.384917824,
+    "generation_bytes_per_second": 432591.03357049276,
+    "upload_bytes_per_second": 4103197.640732556,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.0003491343354,
+    "estimated_full_shards_per_hour": 92.82396560035788,
     "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": 0.9798610010002449,
-      "bytes": 7526317,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7459925380026107,
       "bytes": 508402,
@@ -562,6 +556,12 @@
       "bytes": 5771889,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0350117630005116,
+      "bytes": 6519406,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 107500, "loss": 0.620902725495398, "lr": 0.0003, "grad_norm": 0.12747942690273625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15730833505, "tokens_per_second": 150345.47894324994, "data_wait_s": 1.1673997505567968e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107600, "loss": 0.6109723132103682, "lr": 0.0003, "grad_norm": 0.1507481802760825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15745501220, "tokens_per_second": 149072.38841500712, "data_wait_s": 1.1003001418430358e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107700, "loss": 0.6005883999168873, "lr": 0.0003, "grad_norm": 0.14389146935164826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15760219114, "tokens_per_second": 149446.07856228584, "data_wait_s": 1.5112000255612656e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 107800, "loss": 0.6015025153756142, "lr": 0.0003, "grad_norm": 0.14818491912674067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15774916925, "tokens_per_second": 149166.77545127875, "data_wait_s": 1.4714998542331159e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c6d5683f5297ed026023e6805c02d26631d600318bf5acb64ab033d9e5dcdce
-size 484083
+oid sha256:3b20625f4040964e876b4dff608458a207dcd4734faa8a767b647ab81a40f4c5
+size 486201