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": 74000.39377934652,
-      "required_pages_per_second_upper_estimate": 0.006414166580510409
+      "required_token_positions_per_second": 75446.56077460723,
+      "required_pages_per_second_upper_estimate": 0.006539516670382868
     },
     {
       "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": 9250.049222418314,
-      "required_pages_per_second_upper_estimate": 0.003689982935383084
+      "required_token_positions_per_second": 9430.820096825904,
+      "required_pages_per_second_upper_estimate": 0.0037620951399496986
     },
     {
       "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": 18500.09844483663,
-      "required_pages_per_second_upper_estimate": 0.004279672184600626
+      "required_token_positions_per_second": 18861.64019365181,
+      "required_pages_per_second_upper_estimate": 0.004363308505271559
     },
     {
       "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": 27750.147667254943,
-      "required_pages_per_second_upper_estimate": 0.0016898383991630955
+      "required_token_positions_per_second": 28292.460290477713,
+      "required_pages_per_second_upper_estimate": 0.001722862392622924
     },
     {
       "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": 27750.147667254943,
-      "required_pages_per_second_upper_estimate": 0.002136283184750556
+      "required_token_positions_per_second": 28292.460290477713,
+      "required_pages_per_second_upper_estimate": 0.0021780319117036683
     },
     {
       "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": 27750.147667254943,
-      "required_pages_per_second_upper_estimate": 0.3445639602574586
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28292.460290477713,
+      "required_pages_per_second_upper_estimate": 0.35129766803429124
     }
   ],
   "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": 0.10855237773427007,
-    "producer_p95_seconds": 0.3388429740007268,
-    "gpu_update_seconds": 0.9963190225695455,
-    "producer_batches_per_second": 9.212142754237425,
-    "gpu_batches_per_second": 1.0036945770853207,
-    "producer_headroom": 9.178233064672952,
+    "producer_batch_seconds": 0.09500494727353725,
+    "producer_p95_seconds": 0.30927219500154024,
+    "gpu_update_seconds": 0.9772214829017675,
+    "producer_batches_per_second": 10.525767643666075,
+    "gpu_batches_per_second": 1.0233094723118386,
+    "producer_headroom": 10.286006265422806,
     "sustainable": true,
-    "buffer_seconds": 60.775460376742274,
+    "buffer_seconds": 60.58773193990958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418051.1124307519,
-    "upload_bytes_per_second": 3963764.016978586,
+    "generation_bytes_per_second": 433636.24248698476,
+    "upload_bytes_per_second": 3946292.6944979886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.70403699581068,
+    "estimated_full_shards_per_hour": 93.04824310261876,
     "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": 1.286527730000671,
-      "bytes": 7926367,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.945500746000107,
       "bytes": 979107,
@@ -562,6 +556,12 @@
       "bytes": 7704490,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2197817560008843,
+      "bytes": 5784149,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 137800, "loss": 0.599891185760498, "lr": 0.0003, "grad_norm": 0.13112477463219238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20166656510, "tokens_per_second": 149432.94795432422, "data_wait_s": 1.263399826711975e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137900, "loss": 0.661179369315505, "lr": 0.0003, "grad_norm": 0.15528800027034226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20181428733, "tokens_per_second": 150458.80736417428, "data_wait_s": 1.4265002391766757e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138000, "loss": 0.6163253784179688, "lr": 0.0003, "grad_norm": 0.13840160992919742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20196193884, "tokens_per_second": 151910.37238622457, "data_wait_s": 8.898001397028565e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 138100, "loss": 0.618504099547863, "lr": 0.0003, "grad_norm": 0.17327833131624185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20210930534, "tokens_per_second": 149147.4022133526, "data_wait_s": 1.811800029827282e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc34a02efd9eb982f27770d03d6af32d1f5a8003330110405aa2e6b47fb0d9c4
-size 490437
+oid sha256:1ab4664148a2c31a19c864383b0ff57ea5c275dc7533593d470c97428b36a981
+size 492555