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": 75505.9637153255,
-      "required_pages_per_second_upper_estimate": 0.00654466556672895
+      "required_token_positions_per_second": 75674.71597347544,
+      "required_pages_per_second_upper_estimate": 0.006559292584766699
     },
     {
       "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": 9438.245464415688,
-      "required_pages_per_second_upper_estimate": 0.003765057230100402
+      "required_token_positions_per_second": 9459.33949668443,
+      "required_pages_per_second_upper_estimate": 0.0037734719549562907
     },
     {
       "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": 18876.490928831376,
-      "required_pages_per_second_upper_estimate": 0.004366743961491339
+      "required_token_positions_per_second": 18918.67899336886,
+      "required_pages_per_second_upper_estimate": 0.004376503427737509
     },
     {
       "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": 28314.73639324706,
-      "required_pages_per_second_upper_estimate": 0.0017242188904078987
+      "required_token_positions_per_second": 28378.018490053288,
+      "required_pages_per_second_upper_estimate": 0.0017280724380878954
     },
     {
       "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": 28314.73639324706,
-      "required_pages_per_second_upper_estimate": 0.0021797467877590518
+      "required_token_positions_per_second": 28378.018490053288,
+      "required_pages_per_second_upper_estimate": 0.0021846184187472506
     },
     {
       "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": 28314.73639324706,
-      "required_pages_per_second_upper_estimate": 0.35157426267736647
+      "required_token_positions_per_second": 28378.018490053288,
+      "required_pages_per_second_upper_estimate": 0.3523600145281459
     }
   ],
   "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": 0.3620274261873533,
-    "producer_p95_seconds": 0.3851484909973806,
-    "gpu_update_seconds": 0.9764526717117495,
-    "producer_batches_per_second": 2.762221665168784,
-    "gpu_batches_per_second": 1.0241151762603826,
-    "producer_headroom": 2.697178724814137,
+    "producer_batch_seconds": 0.33417362151533325,
+    "producer_p95_seconds": 0.3994370400032494,
+    "gpu_update_seconds": 0.9742752126858623,
+    "producer_batches_per_second": 2.9924564226985697,
+    "gpu_batches_per_second": 1.0264040252478766,
+    "producer_headroom": 2.915476117677824,
     "sustainable": true,
-    "buffer_seconds": 60.54006564612847,
+    "buffer_seconds": 60.40506318652346,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428462.5434833329,
-    "upload_bytes_per_second": 5692753.399932583,
+    "generation_bytes_per_second": 395220.52181059885,
+    "upload_bytes_per_second": 5676283.9716359945,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.9380877339839,
+    "estimated_full_shards_per_hour": 84.80512371767496,
     "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": 1.1778193410063977,
-      "bytes": 11708390,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.1431241999962367,
       "bytes": 977191,
@@ -562,6 +556,12 @@
       "bytes": 1079259,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1580835760032642,
+      "bytes": 10392878,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -333,3 +333,4 @@
 {"step": 313500, "loss": 0.5657958760857582, "lr": 0.0003, "grad_norm": 0.14890168583214022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45887546884, "tokens_per_second": 148605.22678918854, "data_wait_s": 1.1411997547838837e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 313600, "loss": 0.5694801658391953, "lr": 0.0003, "grad_norm": 0.15211941678975824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45902177731, "tokens_per_second": 149532.32139601294, "data_wait_s": 9.321003744844347e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 313700, "loss": 0.6515887826681137, "lr": 0.0003, "grad_norm": 0.1977031417502658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45916832172, "tokens_per_second": 150912.25244320382, "data_wait_s": 1.0141993698198348e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 313800, "loss": 0.564821919426322, "lr": 0.0003, "grad_norm": 0.152956664917729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45931584558, "tokens_per_second": 151120.3533412845, "data_wait_s": 1.4492994523607194e-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 sha256:94d6afcc40229f79cae1ecad2bfdf5b99ca186ff4c88f8152445435bdf1d9b17
-size 714393
+oid sha256:2718b2ff96c9aa752539b56d5c8ddf5c353988470fdfbdb9800bdaa9c6a4eed3
+size 716511