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": 75813.20649810691,
-      "required_pages_per_second_upper_estimate": 0.006571296592440729
+      "required_token_positions_per_second": 75504.7173614295,
+      "required_pages_per_second_upper_estimate": 0.0065445575359321
     },
     {
       "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": 9476.650812263364,
-      "required_pages_per_second_upper_estimate": 0.0037803776975679606
+      "required_token_positions_per_second": 9438.089670178688,
+      "required_pages_per_second_upper_estimate": 0.003764995081449931
     },
     {
       "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": 18953.301624526728,
-      "required_pages_per_second_upper_estimate": 0.004384512764118385
+      "required_token_positions_per_second": 18876.179340357376,
+      "required_pages_per_second_upper_estimate": 0.004366671880981649
     },
     {
       "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": 28429.952436790096,
-      "required_pages_per_second_upper_estimate": 0.0017312349429678066
+      "required_token_positions_per_second": 28314.269010536063,
+      "required_pages_per_second_upper_estimate": 0.0017241904292529665
     },
     {
       "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": 28429.952436790096,
-      "required_pages_per_second_upper_estimate": 0.002188616437729416
+      "required_token_positions_per_second": 28314.269010536063,
+      "required_pages_per_second_upper_estimate": 0.0021797108073441608
     },
     {
       "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": 28429.952436790096,
-      "required_pages_per_second_upper_estimate": 0.3530048603348783
+      "required_token_positions_per_second": 28314.269010536063,
+      "required_pages_per_second_upper_estimate": 0.35156845934832515
     }
   ],
   "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": 0.1137114780391073,
-    "producer_p95_seconds": 0.30017478999798186,
-    "gpu_update_seconds": 0.972495471509189,
-    "producer_batches_per_second": 8.794186983094908,
-    "gpu_batches_per_second": 1.0282824232056602,
-    "producer_headroom": 8.552307016664855,
+    "producer_batch_seconds": 0.074159519789049,
+    "producer_p95_seconds": 0.24526841199985938,
+    "gpu_update_seconds": 0.9764687899840134,
+    "producer_batches_per_second": 13.484445460873497,
+    "gpu_batches_per_second": 1.0240982715037639,
+    "producer_headroom": 13.167140142784564,
     "sustainable": true,
-    "buffer_seconds": 60.29471923356972,
+    "buffer_seconds": 60.54106497900883,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316082.3565823002,
-    "upload_bytes_per_second": 5790414.661888446,
+    "generation_bytes_per_second": 316102.8025646193,
+    "upload_bytes_per_second": 5741639.056362377,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.82391570188288,
+    "estimated_full_shards_per_hour": 67.82830293373046,
     "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": 1.4506824640011473,
-      "bytes": 7595330,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9329213019991585,
       "bytes": 994205,
@@ -562,6 +556,12 @@
       "bytes": 12184416,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 0.945849695999641,
+      "bytes": 1129028,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -183,3 +183,4 @@
 {"step": 133300, "loss": 0.6020955741405487, "lr": 0.0003, "grad_norm": 0.13759156328564195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19507963850, "tokens_per_second": 150550.9122177826, "data_wait_s": 1.0233001376036555e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133400, "loss": 0.5951715297996998, "lr": 0.0003, "grad_norm": 0.1380262139766603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19522758738, "tokens_per_second": 150925.6095559784, "data_wait_s": 1.2779000826412812e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133500, "loss": 0.6138676777482033, "lr": 0.0003, "grad_norm": 0.13419845753725104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19537352418, "tokens_per_second": 148410.90869555785, "data_wait_s": 1.2267002603039145e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 133600, "loss": 0.5971824824810028, "lr": 0.0003, "grad_norm": 0.13473001862666656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19551872648, "tokens_per_second": 147372.69474714156, "data_wait_s": 1.354100095340982e-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 sha256:68f5d97fe88402739ee7f5cdabfdc4b64c38194fbb2153ab828ccf36da5af19a
-size 394344
+oid sha256:56ba02a6f1c01bd6b6403e0f2815fe4d93d3095590255c9d087593579b4de9ef
+size 396462