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": 75669.90558580896,
-      "required_pages_per_second_upper_estimate": 0.006558875632555584
+      "required_token_positions_per_second": 74643.24945059679,
+      "required_pages_per_second_upper_estimate": 0.00646988768079158
     },
     {
       "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": 9458.73819822612,
-      "required_pages_per_second_upper_estimate": 0.0037732320880110577
+      "required_token_positions_per_second": 9330.406181324599,
+      "required_pages_per_second_upper_estimate": 0.003722038527734402
     },
     {
       "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": 18917.47639645224,
-      "required_pages_per_second_upper_estimate": 0.004376225228105831
+      "required_token_positions_per_second": 18660.812362649198,
+      "required_pages_per_second_upper_estimate": 0.0043168505210182135
     },
     {
       "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": 28376.21459467836,
-      "required_pages_per_second_upper_estimate": 0.00172796259032387
+      "required_token_positions_per_second": 27991.218543973795,
+      "required_pages_per_second_upper_estimate": 0.0017045183507541325
     },
     {
       "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": 28376.21459467836,
-      "required_pages_per_second_upper_estimate": 0.002184479549887787
+      "required_token_positions_per_second": 27991.218543973795,
+      "required_pages_per_second_upper_estimate": 0.0021548414881672707
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28376.21459467836,
-      "required_pages_per_second_upper_estimate": 0.3523376161848388
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27991.218543973795,
+      "required_pages_per_second_upper_estimate": 0.34755725373398305
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11418651941403368,
-    "producer_p95_seconds": 0.26790120800069417,
-    "gpu_update_seconds": 0.9743371480276679,
-    "producer_batches_per_second": 8.757601204867784,
-    "gpu_batches_per_second": 1.0263387801894661,
-    "producer_headroom": 8.532856181514545,
+    "producer_batch_seconds": 0.09595844376568152,
+    "producer_p95_seconds": 0.24993882600028883,
+    "gpu_update_seconds": 0.9877383493171401,
+    "producer_batches_per_second": 10.42117775942547,
+    "gpu_batches_per_second": 1.0124138651610892,
+    "producer_headroom": 10.293396918035407,
     "sustainable": true,
-    "buffer_seconds": 60.40890317771541,
+    "buffer_seconds": 60.252039308345545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333896.9394897944,
-    "upload_bytes_per_second": 5595907.323438782,
+    "generation_bytes_per_second": 333691.8741091176,
+    "upload_bytes_per_second": 5611786.456584306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.64651049156545,
+    "estimated_full_shards_per_hour": 71.60250823454996,
     "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": 0.9189691079991462,
-      "bytes": 2568805,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.252516081000067,
       "bytes": 9866864,
@@ -562,6 +556,12 @@
       "bytes": 1136025,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.5857687079987954,
+      "bytes": 7492427,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 377300, "loss": 0.5944837853312492, "lr": 0.0003, "grad_norm": 0.1480909025327743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55225745724, "tokens_per_second": 149216.04525386522, "data_wait_s": 1.3347998901735991e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377400, "loss": 0.5850543230772018, "lr": 0.0003, "grad_norm": 0.1613736763972774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55240333260, "tokens_per_second": 148100.013608865, "data_wait_s": 1.1454998457338661e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377500, "loss": 0.791090477257967, "lr": 0.0003, "grad_norm": 0.25889471560862726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55255030746, "tokens_per_second": 149188.24000051804, "data_wait_s": 1.0630999895511195e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 377600, "loss": 0.5957522951066494, "lr": 0.0003, "grad_norm": 0.16499344791161374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55269560771, "tokens_per_second": 130324.79640175137, "data_wait_s": 1.1600997822824866e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f169ec25f1ddc8303e1c49cd1a9c2c7cf2ec561c03c1627fb9becad50d7a595
-size 479847
+oid sha256:869d5d9b6f45b3b881b5a9e888921808b4298347b1b21931064192b3df49ebd2
+size 481965