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": 76262.6479749922,
-      "required_pages_per_second_upper_estimate": 0.006610253040558137
+      "required_token_positions_per_second": 76216.92050433328,
+      "required_pages_per_second_upper_estimate": 0.006606289499296637
     },
     {
       "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": 9532.830996874025,
-      "required_pages_per_second_upper_estimate": 0.0038027888131777665
+      "required_token_positions_per_second": 9527.11506304166,
+      "required_pages_per_second_upper_estimate": 0.0038005086417112095
     },
     {
       "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": 19065.66199374805,
-      "required_pages_per_second_upper_estimate": 0.0044105053580627785
+      "required_token_positions_per_second": 19054.23012608332,
+      "required_pages_per_second_upper_estimate": 0.004407860796672543
     },
     {
       "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": 28598.49299062208,
-      "required_pages_per_second_upper_estimate": 0.001741498178432228
+      "required_token_positions_per_second": 28581.34518912498,
+      "required_pages_per_second_upper_estimate": 0.001740453967288616
     },
     {
       "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": 28598.49299062208,
-      "required_pages_per_second_upper_estimate": 0.0022015911561135155
+      "required_token_positions_per_second": 28581.34518912498,
+      "required_pages_per_second_upper_estimate": 0.002200271070886116
     },
     {
       "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": 28598.49299062208,
-      "required_pages_per_second_upper_estimate": 0.35509756994452335
+      "required_token_positions_per_second": 28581.34518912498,
+      "required_pages_per_second_upper_estimate": 0.3548846516399292
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0709266118124745,
-    "producer_p95_seconds": 0.1776282490000085,
-    "gpu_update_seconds": 0.9667642280684594,
-    "producer_batches_per_second": 14.099080365546532,
-    "gpu_batches_per_second": 1.034378363376088,
-    "producer_headroom": 13.630486546072765,
+    "producer_batch_seconds": 0.07477059708589451,
+    "producer_p95_seconds": 0.17951727800027584,
+    "gpu_update_seconds": 0.9673442525903185,
+    "producer_batches_per_second": 13.37424119873252,
+    "gpu_batches_per_second": 1.0337581448612911,
+    "producer_headroom": 12.937495356350555,
     "sustainable": true,
-    "buffer_seconds": 60.906146368312946,
+    "buffer_seconds": 60.94268791319007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356027.861134759,
-    "upload_bytes_per_second": 6240210.697473412,
+    "generation_bytes_per_second": 369886.6426906674,
+    "upload_bytes_per_second": 6249709.309424514,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.39529109508588,
+    "estimated_full_shards_per_hour": 79.36906300106065,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 13377142,
       "pages": 44,
       "shards": 5
+    },
+    {
+      "seconds": 0.9435872209996887,
+      "bytes": 6172084,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -58,3 +58,4 @@
 {"step": 45800, "loss": 0.6263294331729412, "lr": 0.0003, "grad_norm": 0.12379982046226583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6702066400, "tokens_per_second": 150151.61350128875, "data_wait_s": 1.4790000022912864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45900, "loss": 0.6565524749457836, "lr": 0.0003, "grad_norm": 0.13112385998803777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6716776605, "tokens_per_second": 150495.40776296184, "data_wait_s": 1.0531999578233808e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46000, "loss": 0.6058008559048176, "lr": 0.0003, "grad_norm": 0.1322720212095008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6731396222, "tokens_per_second": 149629.9861653259, "data_wait_s": 1.0580999514786527e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 46100, "loss": 0.6462961342185736, "lr": 0.0003, "grad_norm": 0.12934043426875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6745887842, "tokens_per_second": 148537.8203293296, "data_wait_s": 1.0992000170517713e-05, "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:b8f5708824a72dc0376d8622f4210b8835248ce6277ffdf35c04e9ec0dd8a2bb
-size 128028
+oid sha256:aa9e4440416f2aafaa8369ed9b7fd52fe8c112c2cf64a7f2aeef931a3c697e19
+size 130146