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": 75538.37655622078,
-      "required_pages_per_second_upper_estimate": 0.006547475029628159
+      "required_token_positions_per_second": 75850.39805127421,
+      "required_pages_per_second_upper_estimate": 0.006574520262008141
     },
     {
       "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": 9442.297069527598,
-      "required_pages_per_second_upper_estimate": 0.003766673475956438
+      "required_token_positions_per_second": 9481.299756409277,
+      "required_pages_per_second_upper_estimate": 0.0037822322308956745
     },
     {
       "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": 18884.594139055196,
-      "required_pages_per_second_upper_estimate": 0.004368618496564991
+      "required_token_positions_per_second": 18962.599512818553,
+      "required_pages_per_second_upper_estimate": 0.004386663666937454
     },
     {
       "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": 28326.89120858279,
-      "required_pages_per_second_upper_estimate": 0.001724959054890457
+      "required_token_positions_per_second": 28443.89926922783,
+      "required_pages_per_second_upper_estimate": 0.00173208423188989
     },
     {
       "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": 28326.89120858279,
-      "required_pages_per_second_upper_estimate": 0.0021806824990902784
+      "required_token_positions_per_second": 28443.89926922783,
+      "required_pages_per_second_upper_estimate": 0.002189690103497776
     },
     {
       "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": 28326.89120858279,
-      "required_pages_per_second_upper_estimate": 0.3517251848042861
+      "required_token_positions_per_second": 28443.89926922783,
+      "required_pages_per_second_upper_estimate": 0.35317803331670944
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08332393907821256,
-    "producer_p95_seconds": 0.34347726899795816,
-    "gpu_update_seconds": 0.9760336846149537,
-    "producer_batches_per_second": 12.001352925254091,
-    "gpu_batches_per_second": 1.0245548035511716,
-    "producer_headroom": 11.713724716000202,
+    "producer_batch_seconds": 0.12787192751579823,
+    "producer_p95_seconds": 0.35899460300061037,
+    "gpu_update_seconds": 0.9720186300164241,
+    "producer_batches_per_second": 7.820324753268872,
+    "gpu_batches_per_second": 1.028786865929826,
+    "producer_headroom": 7.601501352955939,
     "sustainable": true,
-    "buffer_seconds": 60.514088446127126,
+    "buffer_seconds": 60.26515506101829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394040.9042203102,
-    "upload_bytes_per_second": 2735820.65047645,
+    "generation_bytes_per_second": 394901.1997989347,
+    "upload_bytes_per_second": 2732334.990968654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.55200524289111,
+    "estimated_full_shards_per_hour": 84.73660464740782,
     "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": 0.9558858569980657,
-      "bytes": 1199021,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.896848256001249,
       "bytes": 11337881,
@@ -562,6 +556,12 @@
       "bytes": 5704653,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.143835795999621,
+      "bytes": 1177702,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 221100, "loss": 0.5796709470450878, "lr": 0.0003, "grad_norm": 0.13842052592317736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32362371517, "tokens_per_second": 146255.55999118107, "data_wait_s": 1.6242000128841028e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221200, "loss": 0.5717740878462791, "lr": 0.0003, "grad_norm": 0.13649338577095727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32376948957, "tokens_per_second": 148279.05683170797, "data_wait_s": 1.471799987484701e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221300, "loss": 0.5787149202078581, "lr": 0.0003, "grad_norm": 0.1392222626005347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32391682260, "tokens_per_second": 150249.8014659225, "data_wait_s": 1.381599940941669e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 221400, "loss": 0.5887596476823092, "lr": 0.0003, "grad_norm": 0.15217444645913925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32406219146, "tokens_per_second": 145521.58387926136, "data_wait_s": 1.332699685008265e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6efaa6e9de88998225e7d0454dfb155f5dcdb54756b3da818f624e4bd1de3d6
-size 561114
+oid sha256:1d01c6644c1e3053732cf1f70aed07aa8c35f44885ba02ac4d7b49ffeaaae6d2
+size 563232