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": 75661.76059331131,
-      "required_pages_per_second_upper_estimate": 0.006558169645249183
+      "required_token_positions_per_second": 76081.5264882095,
+      "required_pages_per_second_upper_estimate": 0.0065945538891320245
     },
     {
       "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": 9457.720074163914,
-      "required_pages_per_second_upper_estimate": 0.0037728259431003327
+      "required_token_positions_per_second": 9510.190811026188,
+      "required_pages_per_second_upper_estimate": 0.0037937573045421207
     },
     {
       "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": 18915.440148327827,
-      "required_pages_per_second_upper_estimate": 0.004375754177938992
+      "required_token_positions_per_second": 19020.381622052377,
+      "required_pages_per_second_upper_estimate": 0.004400030540978308
     },
     {
       "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": 28373.16022249174,
-      "required_pages_per_second_upper_estimate": 0.0017277765950827573
+      "required_token_positions_per_second": 28530.572433078567,
+      "required_pages_per_second_upper_estimate": 0.0017373621728294531
     },
     {
       "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": 28373.16022249174,
-      "required_pages_per_second_upper_estimate": 0.0021842444158618173
+      "required_token_positions_per_second": 28530.572433078567,
+      "required_pages_per_second_upper_estimate": 0.0021963624435776792
     },
     {
       "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": 28373.16022249174,
-      "required_pages_per_second_upper_estimate": 0.35229969110460707
+      "required_token_positions_per_second": 28530.572433078567,
+      "required_pages_per_second_upper_estimate": 0.3542542239353163
     }
   ],
   "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": 0.05907964213288608,
-    "producer_p95_seconds": 0.1968358600006468,
-    "gpu_update_seconds": 0.9744420354727741,
-    "producer_batches_per_second": 16.92630428855222,
-    "gpu_batches_per_second": 1.0262283066584108,
-    "producer_headroom": 16.49370240396837,
+    "producer_batch_seconds": 0.17838651551559792,
+    "producer_p95_seconds": 0.268795157000568,
+    "gpu_update_seconds": 0.9690657299236203,
+    "producer_batches_per_second": 5.60580488446483,
+    "gpu_batches_per_second": 1.0319217459880847,
+    "producer_headroom": 5.4323934021733065,
     "sustainable": true,
-    "buffer_seconds": 60.415406199311995,
+    "buffer_seconds": 60.082075255264456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414929.6454920391,
-    "upload_bytes_per_second": 5674773.773016567,
+    "generation_bytes_per_second": 420635.50044554,
+    "upload_bytes_per_second": 5735663.990811707,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.03424285479431,
+    "estimated_full_shards_per_hour": 90.25858650231028,
     "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": 1.0198160460004146,
-      "bytes": 6464312,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.407603981999273,
       "bytes": 11506972,
@@ -562,6 +556,12 @@
       "bytes": 6780113,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2669327539988444,
+      "bytes": 12406942,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 288200, "loss": 0.6017746590077877, "lr": 0.0003, "grad_norm": 0.14384964663836383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42184330783, "tokens_per_second": 150571.6507652425, "data_wait_s": 9.387998943566345e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 288300, "loss": 0.5978507325053215, "lr": 0.0003, "grad_norm": 0.13419214173368446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42198897066, "tokens_per_second": 148901.08860646313, "data_wait_s": 9.408999176230282e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 288400, "loss": 0.571783060207963, "lr": 0.0003, "grad_norm": 0.14598619365887963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42213362594, "tokens_per_second": 148148.08819389716, "data_wait_s": 1.1122998330392875e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 288500, "loss": 0.6058273836970329, "lr": 0.0003, "grad_norm": 0.1596113652324245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42227982364, "tokens_per_second": 150208.51362881463, "data_wait_s": 1.1613003152888268e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77d516c43eca40a2751d174f3ce50f8d17baf5e0cd13bbbbe3189a2cd239d93c
-size 174624
+oid sha256:f0164e69f732732e20d5bcf3b449fa72eaabc8d1dd770d83e34282f1bd650ddd
+size 176742