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": 75818.68456994633,
-      "required_pages_per_second_upper_estimate": 0.006571771417823724
+      "required_token_positions_per_second": 76061.5759847315,
+      "required_pages_per_second_upper_estimate": 0.00659282463005464
     },
     {
       "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": 9477.335571243291,
-      "required_pages_per_second_upper_estimate": 0.003780650858163113
+      "required_token_positions_per_second": 9507.696998091438,
+      "required_pages_per_second_upper_estimate": 0.003792762485276623
     },
     {
       "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": 18954.671142486583,
-      "required_pages_per_second_upper_estimate": 0.004384829577995704
+      "required_token_positions_per_second": 19015.393996182876,
+      "required_pages_per_second_upper_estimate": 0.004398876741512617
     },
     {
       "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": 28432.006713729876,
-      "required_pages_per_second_upper_estimate": 0.001731360037655481
+      "required_token_positions_per_second": 28523.090994274316,
+      "required_pages_per_second_upper_estimate": 0.0017369065924583503
     },
     {
       "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": 28432.006713729876,
-      "required_pages_per_second_upper_estimate": 0.002188774581654843
+      "required_token_positions_per_second": 28523.090994274316,
+      "required_pages_per_second_upper_estimate": 0.00219578650170858
     },
     {
       "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": 28432.006713729876,
-      "required_pages_per_second_upper_estimate": 0.3530303675792478
+      "required_token_positions_per_second": 28523.090994274316,
+      "required_pages_per_second_upper_estimate": 0.3541613295041325
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07419357953909866,
-    "producer_p95_seconds": 0.21402147700064234,
-    "gpu_update_seconds": 0.9724252065067474,
-    "producer_batches_per_second": 13.478255210385397,
-    "gpu_batches_per_second": 1.028356724310253,
-    "producer_headroom": 13.106595106309664,
+    "producer_batch_seconds": 0.0576654713281215,
+    "producer_p95_seconds": 0.19882392499857815,
+    "gpu_update_seconds": 0.9693199101580547,
+    "producer_batches_per_second": 17.341399922146024,
+    "gpu_batches_per_second": 1.0316511499665189,
+    "producer_headroom": 16.809364214549483,
     "sustainable": true,
-    "buffer_seconds": 60.29036280341833,
+    "buffer_seconds": 60.09783442979939,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376185.9580624353,
-    "upload_bytes_per_second": 4093032.6996984086,
+    "generation_bytes_per_second": 375836.6200271178,
+    "upload_bytes_per_second": 4041937.816756535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.72074943928523,
+    "estimated_full_shards_per_hour": 80.64578962907935,
     "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": 0.9038640889998533,
-      "bytes": 6072537,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.318457908000255,
       "bytes": 13323724,
@@ -562,6 +556,12 @@
       "bytes": 13419202,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 0.9201347599992005,
+      "bytes": 838883,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -108,3 +108,4 @@
 {"step": 325800, "loss": 0.584720553830266, "lr": 0.0003, "grad_norm": 0.14197307855976757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47688314798, "tokens_per_second": 149828.13113493362, "data_wait_s": 1.1918999007320963e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325900, "loss": 0.5973014533519745, "lr": 0.0003, "grad_norm": 0.15332335554336202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47703093685, "tokens_per_second": 150954.89316349427, "data_wait_s": 1.146400063589681e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326000, "loss": 0.6318767759948969, "lr": 0.0003, "grad_norm": 0.1609707136957602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47717613767, "tokens_per_second": 148438.42796112073, "data_wait_s": 1.2656000762945041e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 326100, "loss": 0.5768279675394297, "lr": 0.0003, "grad_norm": 0.15366335743806186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47732037859, "tokens_per_second": 134682.01316421214, "data_wait_s": 1.824300125008449e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03a9ef1058596012cc898327f52a1df08728fcce54296561fb4d91215394c949
-size 234711
+oid sha256:7028b885752647948176a39f99e17db01a0660f76e9ce4d0a477226cbf6ea3d4
+size 236829