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": 74929.94868932894,
-      "required_pages_per_second_upper_estimate": 0.006494738044172305
+      "required_token_positions_per_second": 74300.97844211073,
+      "required_pages_per_second_upper_estimate": 0.006440220497253948
     },
     {
       "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": 9366.243586166118,
-      "required_pages_per_second_upper_estimate": 0.0037363346043426354
+      "required_token_positions_per_second": 9287.622305263842,
+      "required_pages_per_second_upper_estimate": 0.0037049713998978146
     },
     {
       "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": 18732.487172332236,
-      "required_pages_per_second_upper_estimate": 0.0043334312268048176
+      "required_token_positions_per_second": 18575.244610527683,
+      "required_pages_per_second_upper_estimate": 0.004297055927505905
     },
     {
       "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": 28098.730758498357,
-      "required_pages_per_second_upper_estimate": 0.0017110652805456825
+      "required_token_positions_per_second": 27862.866915791525,
+      "required_pages_per_second_upper_estimate": 0.0016967024100067816
     },
     {
       "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": 28098.730758498357,
-      "required_pages_per_second_upper_estimate": 0.0021631180760542297
+      "required_token_positions_per_second": 27862.866915791525,
+      "required_pages_per_second_upper_estimate": 0.0021449606245297526
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28098.730758498357,
-      "required_pages_per_second_upper_estimate": 0.34889219561814266
+      "required_token_positions_per_second": 27862.866915791525,
+      "required_pages_per_second_upper_estimate": 0.3459635560772257
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12476180189048591,
-    "producer_p95_seconds": 0.38110617899656063,
-    "gpu_update_seconds": 0.9839590349339166,
-    "producer_batches_per_second": 8.015273784501648,
-    "gpu_batches_per_second": 1.0163024724572611,
-    "producer_headroom": 7.886701057729364,
+    "producer_batch_seconds": 0.4419971969609833,
+    "producer_p95_seconds": 0.37277480000193464,
+    "gpu_update_seconds": 0.9922884132332503,
+    "producer_batches_per_second": 2.262457786781561,
+    "gpu_batches_per_second": 1.0077715174982467,
+    "producer_headroom": 2.2450106472526867,
     "sustainable": true,
-    "buffer_seconds": 60.02150113096891,
+    "buffer_seconds": 60.529593207228274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352670.733277303,
-    "upload_bytes_per_second": 5888387.962433553,
+    "generation_bytes_per_second": 320424.451394031,
+    "upload_bytes_per_second": 5810868.1961958185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.67492960681264,
+    "estimated_full_shards_per_hour": 68.75562816968629,
     "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": 1.292463003002922,
-      "bytes": 13118636,
-      "pages": 23,
-      "shards": 4
-    },
     {
       "seconds": 1.014325338001072,
       "bytes": 5232610,
@@ -562,6 +556,12 @@
       "bytes": 7787197,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9909355400013737,
+      "bytes": 5704571,
+      "pages": 21,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"step": 232900, "loss": 0.5621838551014662, "lr": 0.0003, "grad_norm": 0.1732460337275868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34089887453, "tokens_per_second": 147968.77266469318, "data_wait_s": 1.0024996299762279e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233000, "loss": 0.5900715999305248, "lr": 0.0003, "grad_norm": 0.16285978683707156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34104661909, "tokens_per_second": 149594.95634956044, "data_wait_s": 1.3912002032157034e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233100, "loss": 0.5716875307261944, "lr": 0.0003, "grad_norm": 0.16265405723075357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34119307938, "tokens_per_second": 137583.99401916174, "data_wait_s": 8.281996997538954e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 233200, "loss": 0.5668853744864464, "lr": 0.0003, "grad_norm": 0.14522512206479846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34134025710, "tokens_per_second": 149567.2520319809, "data_wait_s": 1.1172000085934997e-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:bfd713314406ea19332bd0ad297a43bade8c9212ec0df6d55c3dd8617da6579a
-size 812604
+oid sha256:62de7f5dcf08655ebe034ac17f55d25db59e6e0ab7b7009652af34e45201b989
+size 814722