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": 74734.8864154142,
-      "required_pages_per_second_upper_estimate": 0.006477830540650189
+      "required_token_positions_per_second": 74905.2735778395,
+      "required_pages_per_second_upper_estimate": 0.006492599268046898
     },
     {
       "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": 9341.860801926776,
-      "required_pages_per_second_upper_estimate": 0.00372660794715445
+      "required_token_positions_per_second": 9363.159197229937,
+      "required_pages_per_second_upper_estimate": 0.0037351041954802686
     },
     {
       "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": 18683.72160385355,
-      "required_pages_per_second_upper_estimate": 0.004322150170781966
+      "required_token_positions_per_second": 18726.318394459875,
+      "required_pages_per_second_upper_estimate": 0.00433200418861083
     },
     {
       "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": 28025.582405780326,
-      "required_pages_per_second_upper_estimate": 0.001706610929644901
+      "required_token_positions_per_second": 28089.47759168981,
+      "required_pages_per_second_upper_estimate": 0.0017105018112346564
     },
     {
       "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": 28025.582405780326,
-      "required_pages_per_second_upper_estimate": 0.0021574869133743953
+      "required_token_positions_per_second": 28089.47759168981,
+      "required_pages_per_second_upper_estimate": 0.0021624057416589023
     },
     {
       "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": 28025.582405780326,
-      "required_pages_per_second_upper_estimate": 0.347983937889173
+      "required_token_positions_per_second": 28089.47759168981,
+      "required_pages_per_second_upper_estimate": 0.3487773022547377
     }
   ],
   "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": 0.05783258674225067,
-    "producer_p95_seconds": 0.20834068400108663,
-    "gpu_update_seconds": 0.9865272235804652,
-    "producer_batches_per_second": 17.291289501830832,
-    "gpu_batches_per_second": 1.0136567710424018,
-    "producer_headroom": 17.05832782436722,
+    "producer_batch_seconds": 0.0751914435937806,
+    "producer_p95_seconds": 0.1906919030006975,
+    "gpu_update_seconds": 0.9842831683056854,
+    "producer_batches_per_second": 13.29938557108264,
+    "gpu_batches_per_second": 1.0159677948383177,
+    "producer_headroom": 13.090361366424139,
     "sustainable": true,
-    "buffer_seconds": 60.17816063840838,
+    "buffer_seconds": 60.04127326664681,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397308.96389371104,
-    "upload_bytes_per_second": 6119065.179176733,
+    "generation_bytes_per_second": 396849.93212141754,
+    "upload_bytes_per_second": 6190803.877461106,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.25325477226733,
+    "estimated_full_shards_per_hour": 85.1547572396459,
     "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": 0.9811290180000469,
-      "bytes": 1073130,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.3945427510002446,
       "bytes": 16977562,
@@ -562,6 +556,12 @@
       "bytes": 7382695,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9901690320002672,
+      "bytes": 6106536,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -103,3 +103,4 @@
 {"step": 50300, "loss": 0.6121636796742678, "lr": 0.0003, "grad_norm": 0.12666534850508873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7361338036, "tokens_per_second": 148150.24210174807, "data_wait_s": 1.0711000868468545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50400, "loss": 0.6780245490372181, "lr": 0.0003, "grad_norm": 0.19957534603921212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7375921768, "tokens_per_second": 148934.62356921754, "data_wait_s": 8.644999979878776e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 50500, "loss": 0.5936776921153069, "lr": 0.0003, "grad_norm": 0.12213298037925359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7390598106, "tokens_per_second": 150642.64040251233, "data_wait_s": 1.0911999197560363e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 50600, "loss": 0.6595828458666801, "lr": 0.0003, "grad_norm": 0.14864370134429494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7405328376, "tokens_per_second": 151175.29936112068, "data_wait_s": 8.854000043356791e-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 sha256:68221c9265244b4227141275a94a934aeba524ba5b47d99c37a6e17718dcf259
-size 224121
+oid sha256:1c629b70afeec840af66812597f66675103d49e64ebb48916489f32cea3beaf5
+size 226239