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": 75146.32999636677,
-      "required_pages_per_second_upper_estimate": 0.006513493427452935
+      "required_token_positions_per_second": 74776.1732922313,
+      "required_pages_per_second_upper_estimate": 0.006481409182492062
     },
     {
       "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": 9393.291249545846,
-      "required_pages_per_second_upper_estimate": 0.003747124321663414
+      "required_token_positions_per_second": 9347.021661528912,
+      "required_pages_per_second_upper_estimate": 0.0037286666912114694
     },
     {
       "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": 18786.58249909169,
-      "required_pages_per_second_upper_estimate": 0.004345945228605383
+      "required_token_positions_per_second": 18694.043323057824,
+      "required_pages_per_second_upper_estimate": 0.004324537918863821
     },
     {
       "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": 28179.87374863754,
-      "required_pages_per_second_upper_estimate": 0.0017160064629207913
+      "required_token_positions_per_second": 28041.064984586734,
+      "required_pages_per_second_upper_estimate": 0.0017075537374633982
     },
     {
       "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": 28179.87374863754,
-      "required_pages_per_second_upper_estimate": 0.0021693646880533166
+      "required_token_positions_per_second": 28041.064984586734,
+      "required_pages_per_second_upper_estimate": 0.002158678805149434
     },
     {
       "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": 28179.87374863754,
-      "required_pages_per_second_upper_estimate": 0.3498997199875528
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28041.064984586734,
+      "required_pages_per_second_upper_estimate": 0.3481761797010906
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11536691643748043,
-    "producer_p95_seconds": 0.20817290499871888,
-    "gpu_update_seconds": 0.9811257582847314,
-    "producer_batches_per_second": 8.667996258198679,
-    "gpu_batches_per_second": 1.0192373317649572,
-    "producer_headroom": 8.504394401634393,
+    "producer_batch_seconds": 0.08624723141410584,
+    "producer_p95_seconds": 0.218277064999711,
+    "gpu_update_seconds": 0.9859825229604229,
+    "producer_batches_per_second": 11.594575079153772,
+    "gpu_batches_per_second": 1.0142167601485363,
+    "producer_headroom": 11.432048389198082,
     "sustainable": true,
-    "buffer_seconds": 60.82979701365335,
+    "buffer_seconds": 60.1449339005858,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441900.9453105105,
-    "upload_bytes_per_second": 5282003.239084051,
+    "generation_bytes_per_second": 447931.2350796972,
+    "upload_bytes_per_second": 5250166.155069888,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.82165593611228,
+    "estimated_full_shards_per_hour": 96.11561574261844,
     "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": 1.527598386000136,
-      "bytes": 5658593,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.7142034060007063,
       "bytes": 6133193,
@@ -562,6 +556,12 @@
       "bytes": 7694144,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 2.1076598479994573,
+      "bytes": 6179293,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -132,3 +132,4 @@
 {"step": 208200, "loss": 0.5959686599671841, "lr": 0.0003, "grad_norm": 0.17725419329156736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30472798873, "tokens_per_second": 146882.4358199653, "data_wait_s": 2.543499977036845e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208300, "loss": 0.5849227532744408, "lr": 0.0003, "grad_norm": 0.1436915649278191, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30487336663, "tokens_per_second": 146661.76946535383, "data_wait_s": 1.5249999705702066e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208400, "loss": 0.5508072338998318, "lr": 0.0003, "grad_norm": 0.1494632014841731, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30501775990, "tokens_per_second": 146016.4629212084, "data_wait_s": 1.8898999769589864e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 208500, "loss": 0.6042758673429489, "lr": 0.0003, "grad_norm": 0.14450466836031053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30516426206, "tokens_per_second": 148336.9974279627, "data_wait_s": 2.116799987561535e-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:fee8b68193da1c76ab3191a38c574140a668f680bbfdff810ceccff13d68556e
-size 285543
+oid sha256:a55fe733b3888cce5c75b36ad0301d7ef8da0698be817c3a9307a330e657558d
+size 287661