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": 74805.29650126341,
-      "required_pages_per_second_upper_estimate": 0.006483933508438866
+      "required_token_positions_per_second": 75559.00860847537,
+      "required_pages_per_second_upper_estimate": 0.006549263363626129
     },
     {
       "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": 9350.662062657926,
-      "required_pages_per_second_upper_estimate": 0.0037301189016506805
+      "required_token_positions_per_second": 9444.876076059421,
+      "required_pages_per_second_upper_estimate": 0.003767702280221566
     },
     {
       "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": 18701.324125315852,
-      "required_pages_per_second_upper_estimate": 0.004326222204328473
+      "required_token_positions_per_second": 18889.752152118843,
+      "required_pages_per_second_upper_estimate": 0.004369811712109335
     },
     {
       "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": 28051.98618797378,
-      "required_pages_per_second_upper_estimate": 0.0017082187814505434
+      "required_token_positions_per_second": 28334.628228178262,
+      "required_pages_per_second_upper_estimate": 0.0017254301987908167
     },
     {
       "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": 28051.98618797378,
-      "required_pages_per_second_upper_estimate": 0.0021595195496179947
+      "required_token_positions_per_second": 28334.628228178262,
+      "required_pages_per_second_upper_estimate": 0.002181278116276179
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28051.98618797378,
-      "required_pages_per_second_upper_estimate": 0.34831178449624123
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28334.628228178262,
+      "required_pages_per_second_upper_estimate": 0.351821252693523
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10178072232781687,
-    "producer_p95_seconds": 0.32806254000024637,
-    "gpu_update_seconds": 0.9855986600996198,
-    "producer_batches_per_second": 9.825043260934866,
-    "gpu_batches_per_second": 1.0146117689515979,
-    "producer_headroom": 9.683549473398203,
+    "producer_batch_seconds": 0.19624237472669392,
+    "producer_p95_seconds": 0.3505261579994112,
+    "gpu_update_seconds": 0.975767170027824,
+    "producer_batches_per_second": 5.095739395697267,
+    "gpu_batches_per_second": 1.024834643669642,
+    "producer_headroom": 4.972255209338816,
     "sustainable": true,
-    "buffer_seconds": 60.12151826607681,
+    "buffer_seconds": 60.49756454172509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447796.1585597014,
-    "upload_bytes_per_second": 6079457.666785813,
+    "generation_bytes_per_second": 491554.28808109014,
+    "upload_bytes_per_second": 6002354.254787021,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.08663146584779,
+    "estimated_full_shards_per_hour": 105.47610742401626,
     "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": 1.1060435590006819,
-      "bytes": 11753109,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.0484200940009032,
       "bytes": 6644677,
@@ -562,6 +556,12 @@
       "bytes": 7913979,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9844400470028631,
+      "bytes": 5585376,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 114200, "loss": 0.5817533507943153, "lr": 0.0003, "grad_norm": 0.151245004184825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16711235399, "tokens_per_second": 147684.33645460388, "data_wait_s": 1.0333998943679035e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114300, "loss": 0.6063061691820621, "lr": 0.0003, "grad_norm": 0.1685130957539429, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16725834847, "tokens_per_second": 148726.3324610475, "data_wait_s": 1.841800258262083e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114400, "loss": 0.6041073054075241, "lr": 0.0003, "grad_norm": 0.15474263024281182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16740523488, "tokens_per_second": 150310.6691414158, "data_wait_s": 1.647500175749883e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 114500, "loss": 0.6019764691591263, "lr": 0.0003, "grad_norm": 0.13809368823880785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16755176941, "tokens_per_second": 149165.51744536447, "data_wait_s": 2.0373001461848617e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6fdcb6cfdea166978edfa22c6042294a0751fbbe9ef209f05095932e7600dcb
-size 626772
+oid sha256:97ea84b6bdaa23ef1a08e8283a88fce02e93614dcb27f680d960bfb5391cbf44
+size 628890