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": 76178.53424683424,
-      "required_pages_per_second_upper_estimate": 0.0066029622757857
+      "required_token_positions_per_second": 76096.7460485454,
+      "required_pages_per_second_upper_estimate": 0.006595873082048317
     },
     {
       "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": 9522.31678085428,
-      "required_pages_per_second_upper_estimate": 0.003798594535205952
+      "required_token_positions_per_second": 9512.093256068176,
+      "required_pages_per_second_upper_estimate": 0.0037945162183134575
     },
     {
       "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": 19044.63356170856,
-      "required_pages_per_second_upper_estimate": 0.004405640800398207
+      "required_token_positions_per_second": 19024.18651213635,
+      "required_pages_per_second_upper_estimate": 0.0044009107353354575
     },
     {
       "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": 28566.95034256284,
-      "required_pages_per_second_upper_estimate": 0.0017395773966569048
+      "required_token_positions_per_second": 28536.279768204527,
+      "required_pages_per_second_upper_estimate": 0.001737709719594551
     },
     {
       "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": 28566.95034256284,
-      "required_pages_per_second_upper_estimate": 0.002199162915750277
+      "required_token_positions_per_second": 28536.279768204527,
+      "required_pages_per_second_upper_estimate": 0.0021968018100345906
     },
     {
       "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": 28566.95034256284,
-      "required_pages_per_second_upper_estimate": 0.3547059158220798
+      "required_token_positions_per_second": 28536.279768204527,
+      "required_pages_per_second_upper_estimate": 0.3543250899363588
     }
   ],
   "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": 0.3942001434846816,
-    "producer_p95_seconds": 0.39259900400065817,
-    "gpu_update_seconds": 0.9678316960143392,
-    "producer_batches_per_second": 2.5367824353388637,
-    "gpu_batches_per_second": 1.0332374979225565,
-    "producer_headroom": 2.455178446813398,
+    "producer_batch_seconds": 0.09744620453903963,
+    "producer_p95_seconds": 0.3641540780008654,
+    "gpu_update_seconds": 0.9688719140890166,
+    "producer_batches_per_second": 10.262072337557,
+    "gpu_batches_per_second": 1.0321281744865642,
+    "producer_headroom": 9.942633668208801,
     "sustainable": true,
-    "buffer_seconds": 60.00556515288903,
+    "buffer_seconds": 60.07005867351903,
     "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": 370352.960672214,
-    "upload_bytes_per_second": 5689682.917070215,
+    "generation_bytes_per_second": 370700.4240870371,
+    "upload_bytes_per_second": 5770713.773875045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.4691239845735,
+    "estimated_full_shards_per_hour": 79.54368154486022,
     "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": 0.9514743170002475,
-      "bytes": 822769,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 4.258936992002418,
       "bytes": 6104340,
@@ -562,6 +556,12 @@
       "bytes": 6652309,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.870253834000323,
+      "bytes": 6235033,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -362,3 +362,4 @@
 {"step": 76200, "loss": 0.6052996497601271, "lr": 0.0003, "grad_norm": 0.14071265434358726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11151709472, "tokens_per_second": 150523.14569166646, "data_wait_s": 1.5386001905426383e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 76300, "loss": 0.6632398217916489, "lr": 0.0003, "grad_norm": 0.13184768924390872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11166301057, "tokens_per_second": 149580.3164920321, "data_wait_s": 9.905998012982309e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 76400, "loss": 0.6222759839147329, "lr": 0.0003, "grad_norm": 0.14488445235744885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11181065755, "tokens_per_second": 150801.17515677874, "data_wait_s": 1.275399699807167e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 76500, "loss": 0.5772983878850937, "lr": 0.0003, "grad_norm": 0.13265661268329665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11195825026, "tokens_per_second": 151144.5666849034, "data_wait_s": 8.283997885882854e-06, "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 sha256:3523b64514e0cbafb99e0426e8fbae05bf2c3ff8dbcf8abbb4df6b2f13b5813e
-size 776598
+oid sha256:bba2bab506e88b521ccde9cd62c90785caab1c2417a7968c695428df261e3e2f
+size 778716