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": 75694.28582196924,
-      "required_pages_per_second_upper_estimate": 0.006560988849634814
+      "required_token_positions_per_second": 75139.94819584527,
+      "required_pages_per_second_upper_estimate": 0.0065129402691582565
     },
     {
       "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": 9461.785727746155,
-      "required_pages_per_second_upper_estimate": 0.003774447793101227
+      "required_token_positions_per_second": 9392.493524480658,
+      "required_pages_per_second_upper_estimate": 0.003746806097207858
     },
     {
       "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": 18923.57145549231,
-      "required_pages_per_second_upper_estimate": 0.004377635212745376
+      "required_token_positions_per_second": 18784.987048961317,
+      "required_pages_per_second_upper_estimate": 0.004345576149296684
     },
     {
       "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": 28385.35718323847,
-      "required_pages_per_second_upper_estimate": 0.0017285193259997257
+      "required_token_positions_per_second": 28177.48057344197,
+      "required_pages_per_second_upper_estimate": 0.001715860731107401
     },
     {
       "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": 28385.35718323847,
-      "required_pages_per_second_upper_estimate": 0.002185183371663449
+      "required_token_positions_per_second": 28177.48057344197,
+      "required_pages_per_second_upper_estimate": 0.002169180454802029
     },
     {
       "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": 28385.35718323847,
-      "required_pages_per_second_upper_estimate": 0.35245113653647975
+      "required_token_positions_per_second": 28177.48057344197,
+      "required_pages_per_second_upper_estimate": 0.349870004761066
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.043375454109380485,
-    "producer_p95_seconds": 0.13137220800012983,
-    "gpu_update_seconds": 0.9740233255308876,
-    "producer_batches_per_second": 23.054513676751053,
-    "gpu_batches_per_second": 1.0266694583057896,
-    "producer_headroom": 22.45563407992639,
+    "producer_batch_seconds": 0.49918478921874687,
+    "producer_p95_seconds": 0.13290486899995813,
+    "gpu_update_seconds": 0.9812090874461991,
+    "producer_batches_per_second": 2.0032661683563275,
+    "gpu_batches_per_second": 1.0191507730556268,
+    "producer_headroom": 1.9656229689647557,
     "sustainable": true,
-    "buffer_seconds": 60.38944618291503,
+    "buffer_seconds": 60.83496342166434,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 137,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 667219.6616544735,
-    "upload_bytes_per_second": 3205372.6831152313,
+    "generation_bytes_per_second": 400645.60452153103,
+    "upload_bytes_per_second": 4119802.729647555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.58490365612819,
+    "estimated_full_shards_per_hour": 85.96922017797897,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 2489087,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.2741382629999407,
+      "bytes": 11572130,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -5,3 +5,4 @@
 {"step": 82400, "loss": 0.6455158451572061, "lr": 0.0003, "grad_norm": 0.15663461873576925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12058784343, "tokens_per_second": 148979.85896397586, "data_wait_s": 1.4244000112739741e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 82500, "loss": 0.6499202214181423, "lr": 0.0003, "grad_norm": 0.13765951160687467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12073486506, "tokens_per_second": 149389.11709760514, "data_wait_s": 8.5680001120636e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 82600, "loss": 0.5864890031516552, "lr": 0.0003, "grad_norm": 0.12823567639076558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12088108413, "tokens_per_second": 147056.68414229367, "data_wait_s": 7.895999942775234e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 82700, "loss": 0.6107049584388733, "lr": 0.0003, "grad_norm": 0.1509346911072232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12102595000, "tokens_per_second": 147294.77227946167, "data_wait_s": 9.600999874237459e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:85a67b34edd62eff374beb840aefc1e29881216eb33fba8e289e90a8788feea6
-size 14991
+oid sha256:6bb2d17c949943fb232f152cde4e41482262329b4ee5b7f5115f7acf5bc0ed8b
+size 17109