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": 76920.42199590697,
-      "required_pages_per_second_upper_estimate": 0.0053215332736456155
+      "required_token_positions_per_second": 77225.93846131285,
+      "required_pages_per_second_upper_estimate": 0.005342669611618261
     },
     {
       "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": 9615.052749488372,
-      "required_pages_per_second_upper_estimate": 0.004025621801077333
+      "required_token_positions_per_second": 9653.242307664106,
+      "required_pages_per_second_upper_estimate": 0.0040416109715968535
     },
     {
       "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": 19230.105498976744,
-      "required_pages_per_second_upper_estimate": 0.004005123843451698
+      "required_token_positions_per_second": 19306.484615328212,
+      "required_pages_per_second_upper_estimate": 0.004021031599134959
     },
     {
       "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": 28845.158248465115,
-      "required_pages_per_second_upper_estimate": 0.001729223192989069
+      "required_token_positions_per_second": 28959.72692299232,
+      "required_pages_per_second_upper_estimate": 0.001736091410090738
     },
     {
       "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": 28845.158248465115,
-      "required_pages_per_second_upper_estimate": 0.0022425841773757436
+      "required_token_positions_per_second": 28959.72692299232,
+      "required_pages_per_second_upper_estimate": 0.002251491387885892
     },
     {
       "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": 28845.158248465115,
-      "required_pages_per_second_upper_estimate": 0.36999471849341486
+      "required_token_positions_per_second": 28959.72692299232,
+      "required_pages_per_second_upper_estimate": 0.37146428243599133
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.16569302892170867,
+    "producer_batch_seconds": 0.3068276530545404,
     "producer_p95_seconds": 0.30582650799988187,
-    "gpu_update_seconds": 0.958497081619276,
-    "producer_batches_per_second": 6.035256923648299,
-    "gpu_batches_per_second": 1.0432999945191375,
-    "producer_headroom": 5.784776148139424,
+    "gpu_update_seconds": 0.9547051349454927,
+    "producer_batches_per_second": 3.2591586515907816,
+    "gpu_batches_per_second": 1.0474438267864699,
+    "producer_headroom": 3.111535500275747,
     "sustainable": true,
-    "buffer_seconds": 60.38531614201439,
+    "buffer_seconds": 60.14642350156604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363776.9184328341,
-    "upload_bytes_per_second": 5728231.695926,
+    "generation_bytes_per_second": 376490.1481785479,
+    "upload_bytes_per_second": 5789703.918124365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.05805840243117,
+    "estimated_full_shards_per_hour": 80.7860215570195,
     "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": 1.2072375589996227,
-      "bytes": 11138063,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9064097570008016,
       "bytes": 1171302,
@@ -562,6 +556,12 @@
       "bytes": 6071704,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.4701623389992164,
+      "bytes": 17195357,
+      "pages": 39,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -289,3 +289,4 @@
 {"step": 586000, "loss": 0.5877169370651245, "lr": 0.0003, "grad_norm": 0.142617230845834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85772972029, "tokens_per_second": 150807.67422692457, "data_wait_s": 1.2798998795915395e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586100, "loss": 0.6081873774528503, "lr": 0.0003, "grad_norm": 0.18081899728163095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85787713849, "tokens_per_second": 153397.38388069393, "data_wait_s": 1.221899947267957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586200, "loss": 0.6391074657440186, "lr": 0.0003, "grad_norm": 0.18716523192384205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85802520527, "tokens_per_second": 153722.66691978186, "data_wait_s": 1.5544999769190326e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 586300, "loss": 0.5808490514755249, "lr": 0.0003, "grad_norm": 0.17973211195187186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85817229576, "tokens_per_second": 152600.74465313466, "data_wait_s": 1.2349999451544136e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6475721ea231484fa2cae00ff036e741f295b1bc5f4037467916c9f07037011
-size 621201
+oid sha256:bd127151669398d70fbf13b89de49c0b91a4dc2c9f221ef792d0a4a34a7cdf28
+size 623319