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": 75554.04480526525,
-      "required_pages_per_second_upper_estimate": 0.0065488331137445194
+      "required_token_positions_per_second": 75677.75585753718,
+      "required_pages_per_second_upper_estimate": 0.0065595560742126545
     },
     {
       "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": 9444.255600658156,
-      "required_pages_per_second_upper_estimate": 0.0037674547633070672
+      "required_token_positions_per_second": 9459.719482192148,
+      "required_pages_per_second_upper_estimate": 0.003773623536856609
     },
     {
       "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": 18888.511201316313,
-      "required_pages_per_second_upper_estimate": 0.004369524639981159
+      "required_token_positions_per_second": 18919.438964384295,
+      "required_pages_per_second_upper_estimate": 0.004376679233656896
     },
     {
       "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": 28332.766801974467,
-      "required_pages_per_second_upper_estimate": 0.001725316847701154
+      "required_token_positions_per_second": 28379.158446576443,
+      "required_pages_per_second_upper_estimate": 0.001728141855459266
     },
     {
       "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": 28332.766801974467,
-      "required_pages_per_second_upper_estimate": 0.0021811348185342533
+      "required_token_positions_per_second": 28379.158446576443,
+      "required_pages_per_second_upper_estimate": 0.002184706175756005
     },
     {
       "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": 28332.766801974467,
-      "required_pages_per_second_upper_estimate": 0.35179814000986465
+      "required_token_positions_per_second": 28379.158446576443,
+      "required_pages_per_second_upper_estimate": 0.35237416897297447
     }
   ],
   "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": 0.08784717199998227,
-    "producer_p95_seconds": 0.3031944449976436,
-    "gpu_update_seconds": 0.9758312766712658,
-    "producer_batches_per_second": 11.383405717377013,
-    "gpu_batches_per_second": 1.0247673177797478,
-    "producer_headroom": 11.108283334054997,
+    "producer_batch_seconds": 0.14095146684374527,
+    "producer_p95_seconds": 0.3730686790004256,
+    "gpu_update_seconds": 0.9742360772271369,
+    "producer_batches_per_second": 7.094640604972002,
+    "gpu_batches_per_second": 1.026445256314252,
+    "producer_headroom": 6.911854832324285,
     "sustainable": true,
-    "buffer_seconds": 60.50153915361848,
+    "buffer_seconds": 60.40263678808249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285362.7176354108,
-    "upload_bytes_per_second": 2848091.342899169,
+    "generation_bytes_per_second": 286750.25283846725,
+    "upload_bytes_per_second": 2915932.1298202733,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.23219630047553,
+    "estimated_full_shards_per_hour": 61.52992905488504,
     "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": 2.7144779629998084,
-      "bytes": 851325,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.4230423680019157,
       "bytes": 11644215,
@@ -562,6 +556,12 @@
       "bytes": 6132224,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.237680772999738,
+      "bytes": 6534213,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 224000, "loss": 0.5858014151453972, "lr": 0.0003, "grad_norm": 0.15248701402295245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32786845177, "tokens_per_second": 148408.81860517786, "data_wait_s": 1.1183001333847642e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 224100, "loss": 0.561779797077179, "lr": 0.0003, "grad_norm": 0.1768840042951018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32801602590, "tokens_per_second": 150094.3845559986, "data_wait_s": 9.258998034056276e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 224200, "loss": 0.5719302129000425, "lr": 0.0003, "grad_norm": 0.1514180616638548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32816282505, "tokens_per_second": 149639.2044351578, "data_wait_s": 9.969000529963523e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 224300, "loss": 0.5948855951428413, "lr": 0.0003, "grad_norm": 0.17108263573307095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32830951308, "tokens_per_second": 131888.25326792768, "data_wait_s": 2.1203999494900927e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b08666985bacc32cf467ec167821769f616a07bf990ab1067a249360bca220b8
-size 622536
+oid sha256:e49b3f98f9875b9ce6af72aa16747c181522ea8e63eaa304d5bd3b72ec6422b3
+size 624654