Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77436.3696287167,
-      "required_pages_per_second_upper_estimate": 0.005357227728046837
+      "required_token_positions_per_second": 77453.817419227,
+      "required_pages_per_second_upper_estimate": 0.005358434806678792
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9679.546203589587,
-      "required_pages_per_second_upper_estimate": 0.004052623863533044
+      "required_token_positions_per_second": 9681.727177403374,
+      "required_pages_per_second_upper_estimate": 0.004053536991724964
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19359.092407179174,
-      "required_pages_per_second_upper_estimate": 0.004031988414816747
+      "required_token_positions_per_second": 19363.45435480675,
+      "required_pages_per_second_upper_estimate": 0.004032896893475277
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29038.63861076876,
-      "required_pages_per_second_upper_estimate": 0.001740822045281765
+      "required_token_positions_per_second": 29045.181532210125,
+      "required_pages_per_second_upper_estimate": 0.0017412142834317122
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29038.63861076876,
-      "required_pages_per_second_upper_estimate": 0.0022576264245146777
+      "required_token_positions_per_second": 29045.181532210125,
+      "required_pages_per_second_upper_estimate": 0.002258135107877474
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29038.63861076876,
-      "required_pages_per_second_upper_estimate": 0.3724764768380185
+      "required_token_positions_per_second": 29045.181532210125,
+      "required_pages_per_second_upper_estimate": 0.3725604024090266
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08800965217963608,
-    "producer_p95_seconds": 0.21772737200080883,
-    "gpu_update_seconds": 0.9521107504587679,
-    "producer_batches_per_second": 11.362390092837826,
-    "gpu_batches_per_second": 1.0502979821603284,
-    "producer_headroom": 10.818253758297093,
+    "producer_batch_seconds": 0.2514586477966674,
+    "producer_p95_seconds": 0.21011276699937298,
+    "gpu_update_seconds": 0.9518962713088677,
+    "producer_batches_per_second": 3.9767970151840335,
+    "gpu_batches_per_second": 1.0505346329647758,
+    "producer_headroom": 3.785498250505916,
     "sustainable": true,
-    "buffer_seconds": 60.93508802936115,
+    "buffer_seconds": 60.92136136376753,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493315.19661256357,
-    "upload_bytes_per_second": 5372182.101540501,
+    "generation_bytes_per_second": 460978.5619639881,
+    "upload_bytes_per_second": 5327136.264896274,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.85395740301782,
+    "estimated_full_shards_per_hour": 98.91526836576207,
     "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": 1.216849121999985,
-      "bytes": 5739383,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.4537456819998624,
       "bytes": 13399952,
@@ -562,6 +556,12 @@
       "bytes": 7707081,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0018968719996337,
+      "bytes": 927002,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 897300, "loss": 0.4963752031326294, "lr": 3e-05, "grad_norm": 0.3112137685009082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131339475719, "tokens_per_second": 152511.61751028922, "data_wait_s": 1.7022000974975526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897400, "loss": 0.5102488994598389, "lr": 3e-05, "grad_norm": 0.3118720297717996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131354045000, "tokens_per_second": 152348.96066376314, "data_wait_s": 2.3840999347157776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897500, "loss": 0.5129820704460144, "lr": 3e-05, "grad_norm": 0.32767872062354386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131368784009, "tokens_per_second": 154032.81284392608, "data_wait_s": 1.7063999621313997e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 897600, "loss": 0.5346129536628723, "lr": 3e-05, "grad_norm": 0.316646003153802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131383490026, "tokens_per_second": 153774.59404020943, "data_wait_s": 1.1744999937945977e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:597d11f9704da31cbfd3b8b41308b8ae4c1c6e983cd569c04cbbefc32e0476c1
-size 266481
+oid sha256:77ed510cedfbe09a6363b94316c3a63a720c4ceef162c140aa397363d24e251b
+size 268599