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": 75664.3309739795,
-      "required_pages_per_second_upper_estimate": 0.006558392439330931
+      "required_token_positions_per_second": 75042.10504141057,
+      "required_pages_per_second_upper_estimate": 0.006504459472513065
     },
     {
       "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": 9458.041371747437,
-      "required_pages_per_second_upper_estimate": 0.0037729541135102266
+      "required_token_positions_per_second": 9380.263130176321,
+      "required_pages_per_second_upper_estimate": 0.0037419272100591675
     },
     {
       "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": 18916.082743494873,
-      "required_pages_per_second_upper_estimate": 0.0043759028310216
+      "required_token_positions_per_second": 18760.526260352643,
+      "required_pages_per_second_upper_estimate": 0.004339917576347244
     },
     {
       "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": 28374.12411524231,
-      "required_pages_per_second_upper_estimate": 0.0017278352910941124
+      "required_token_positions_per_second": 28140.789390528964,
+      "required_pages_per_second_upper_estimate": 0.0017136264305717581
     },
     {
       "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": 28374.12411524231,
-      "required_pages_per_second_upper_estimate": 0.002184318618993982
+      "required_token_positions_per_second": 28140.789390528964,
+      "required_pages_per_second_upper_estimate": 0.0021663558659736908
     },
     {
       "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": 28374.12411524231,
-      "required_pages_per_second_upper_estimate": 0.3523116594266276
+      "required_token_positions_per_second": 28140.789390528964,
+      "required_pages_per_second_upper_estimate": 0.34941442306677634
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.31745338859417416,
-    "producer_p95_seconds": 0.4265484959978494,
-    "gpu_update_seconds": 0.9744089328610415,
-    "producer_batches_per_second": 3.1500687531749088,
-    "gpu_batches_per_second": 1.0262631696774562,
-    "producer_headroom": 3.0694551322200745,
+    "producer_batch_seconds": 0.14230400361725515,
+    "producer_p95_seconds": 0.4531578130045091,
+    "gpu_update_seconds": 0.9824884304526718,
+    "producer_batches_per_second": 7.027209175994993,
+    "gpu_batches_per_second": 1.0178236903403126,
+    "producer_headroom": 6.904151713785933,
     "sustainable": true,
-    "buffer_seconds": 60.41335383738457,
+    "buffer_seconds": 60.91428268806565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 503307.7915602362,
-    "upload_bytes_per_second": 5762091.1431020945,
+    "generation_bytes_per_second": 500792.6854429732,
+    "upload_bytes_per_second": 5837003.744012647,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.99813566308322,
+    "estimated_full_shards_per_hour": 107.45845243899248,
     "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": 1.0604035779979313,
-      "bytes": 6119324,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0826924209977733,
       "bytes": 902486,
@@ -562,6 +556,12 @@
       "bytes": 1110010,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.240684771000815,
+      "bytes": 12447735,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -422,3 +422,4 @@
 {"step": 157200, "loss": 0.6465106811374426, "lr": 0.0003, "grad_norm": 0.15757927262753454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23006794010, "tokens_per_second": 148872.62405405362, "data_wait_s": 1.040300412569195e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157300, "loss": 0.587610375136137, "lr": 0.0003, "grad_norm": 0.15463878370965745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23021545116, "tokens_per_second": 149662.42141581018, "data_wait_s": 1.5237004845403135e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157400, "loss": 0.6482329815626144, "lr": 0.0003, "grad_norm": 0.1487480766401677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23036282419, "tokens_per_second": 148764.09706020146, "data_wait_s": 1.262600562768057e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 157500, "loss": 0.5908825155347586, "lr": 0.0003, "grad_norm": 0.13685302812500091, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23051116604, "tokens_per_second": 150471.47986804944, "data_wait_s": 1.4163997548166662e-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 sha256:21f82ed77bc2a7e44126477d1cb5dd0907faa1258f35b542bae7d8b3e3b073c1
-size 904461
+oid sha256:62623a7b29c72ff41c54e1bcbc5831b294a07266e55274dd010c387457f6213b
+size 906579