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": 75380.93356734431,
-      "required_pages_per_second_upper_estimate": 0.0065338282703879536
+      "required_token_positions_per_second": 75365.36844393174,
+      "required_pages_per_second_upper_estimate": 0.006532479124939997
     },
     {
       "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": 9422.61669591804,
-      "required_pages_per_second_upper_estimate": 0.0037588226806757777
+      "required_token_positions_per_second": 9420.671055491468,
+      "required_pages_per_second_upper_estimate": 0.0037580465356197015
     },
     {
       "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": 18845.23339183608,
-      "required_pages_per_second_upper_estimate": 0.0043595130804213395
+      "required_token_positions_per_second": 18841.342110982936,
+      "required_pages_per_second_upper_estimate": 0.004358612901080168
     },
     {
       "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": 28267.850087754116,
-      "required_pages_per_second_upper_estimate": 0.0017213637603968118
+      "required_token_positions_per_second": 28262.013166474404,
+      "required_pages_per_second_upper_estimate": 0.0017210083225148367
     },
     {
       "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": 28267.850087754116,
-      "required_pages_per_second_upper_estimate": 0.0021761373501726043
+      "required_token_positions_per_second": 28262.013166474404,
+      "required_pages_per_second_upper_estimate": 0.0021756880078148597
     },
     {
       "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": 28267.850087754116,
-      "required_pages_per_second_upper_estimate": 0.3509920916815143
+      "required_token_positions_per_second": 28262.013166474404,
+      "required_pages_per_second_upper_estimate": 0.3509196166541391
     }
   ],
   "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": 0.10046571888278777,
-    "producer_p95_seconds": 0.40375192200008314,
-    "gpu_update_seconds": 0.97807225926875,
-    "producer_batches_per_second": 9.953644000364829,
-    "gpu_batches_per_second": 1.0224193463452733,
-    "producer_headroom": 9.735383075393667,
+    "producer_batch_seconds": 0.12519691132803246,
+    "producer_p95_seconds": 0.4236795250035357,
+    "gpu_update_seconds": 0.9782742594146558,
+    "producer_batches_per_second": 7.987417496106336,
+    "gpu_batches_per_second": 1.0222082308475986,
+    "producer_headroom": 7.8138849356390905,
     "sustainable": true,
-    "buffer_seconds": 60.6404800746625,
+    "buffer_seconds": 60.653004083708666,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288885.32861998497,
-    "upload_bytes_per_second": 6382380.876446889,
+    "generation_bytes_per_second": 329239.1397660883,
+    "upload_bytes_per_second": 6379819.596223703,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.98806661557829,
+    "estimated_full_shards_per_hour": 70.64705509888636,
     "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": 0.9612779050003155,
-      "bytes": 6227883,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9593730830019922,
       "bytes": 6813129,
@@ -562,6 +556,12 @@
       "bytes": 5813086,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.007629627994902,
+      "bytes": 6351245,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 78900, "loss": 0.6395603455603123, "lr": 0.0003, "grad_norm": 0.1199310508753388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11546839230, "tokens_per_second": 149588.12416878866, "data_wait_s": 8.640003215987235e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 79000, "loss": 0.611991485580802, "lr": 0.0003, "grad_norm": 0.12761705911848154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11561471963, "tokens_per_second": 148535.74278928043, "data_wait_s": 1.1676995200105011e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 79100, "loss": 0.6135145165026188, "lr": 0.0003, "grad_norm": 0.14163863155358805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11575914522, "tokens_per_second": 147067.65504461018, "data_wait_s": 1.4654993719886988e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 79200, "loss": 0.5888105966150761, "lr": 0.0003, "grad_norm": 0.12793456312188425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11590374637, "tokens_per_second": 146819.10636228032, "data_wait_s": 1.6117999621201307e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb4c652c3d282fbbae295d5182a8f3acc5216240a118b70aba0525b11e01a6f8
-size 833784
+oid sha256:ed84f9d46c2899db705383055970572cda3bc4fc0c3554b9ea222d970883ff5d
+size 835902