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": 77080.28555101689,
-      "required_pages_per_second_upper_estimate": 0.005332593005322672
+      "required_token_positions_per_second": 76621.60222424015,
+      "required_pages_per_second_upper_estimate": 0.005300860228484309
     },
     {
       "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": 9635.035693877111,
-      "required_pages_per_second_upper_estimate": 0.004033988242601568
+      "required_token_positions_per_second": 9577.700278030019,
+      "required_pages_per_second_upper_estimate": 0.004009983101286023
     },
     {
       "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": 19270.071387754222,
-      "required_pages_per_second_upper_estimate": 0.004013447684112684
+      "required_token_positions_per_second": 19155.400556060038,
+      "required_pages_per_second_upper_estimate": 0.003989564773943978
     },
     {
       "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": 28905.107081631333,
-      "required_pages_per_second_upper_estimate": 0.0017328170340008156
+      "required_token_positions_per_second": 28733.100834090055,
+      "required_pages_per_second_upper_estimate": 0.0017225055220990487
     },
     {
       "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": 28905.107081631333,
-      "required_pages_per_second_upper_estimate": 0.002247244935467367
+      "required_token_positions_per_second": 28733.100834090055,
+      "required_pages_per_second_upper_estimate": 0.0022338722062965584
     },
     {
       "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": 28905.107081631333,
-      "required_pages_per_second_upper_estimate": 0.370763677757229
+      "required_token_positions_per_second": 28733.100834090055,
+      "required_pages_per_second_upper_estimate": 0.368557366299689
     }
   ],
   "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": 0.1281787463906312,
-    "producer_p95_seconds": 0.29596936899906723,
-    "gpu_update_seconds": 0.9565091705738672,
-    "producer_batches_per_second": 7.801605399950234,
-    "gpu_batches_per_second": 1.0454682827557629,
-    "producer_headroom": 7.462307110251002,
+    "producer_batch_seconds": 0.08208887971852619,
+    "producer_p95_seconds": 0.2956290830006765,
+    "gpu_update_seconds": 0.962235164232513,
+    "producer_batches_per_second": 12.181917982422112,
+    "gpu_batches_per_second": 1.0392469919737433,
+    "producer_headroom": 11.721869850482944,
     "sustainable": true,
-    "buffer_seconds": 60.260077746153634,
+    "buffer_seconds": 60.62081534664832,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316812.54554192885,
-    "upload_bytes_per_second": 6215063.730604213,
+    "generation_bytes_per_second": 332490.01122666814,
+    "upload_bytes_per_second": 6060519.67151889,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.98059725469017,
+    "estimated_full_shards_per_hour": 71.34461643791231,
     "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": 1.1265610020018357,
-      "bytes": 11819263,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0389532609988237,
       "bytes": 7773400,
@@ -562,6 +556,12 @@
       "bytes": 8174651,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.1033362929993018,
+      "bytes": 798343,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 510700, "loss": 0.5512087345123291, "lr": 0.0003, "grad_norm": 0.17483065358041508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74749215150, "tokens_per_second": 150938.28669991807, "data_wait_s": 1.5644000086467713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510800, "loss": 0.5592138767242432, "lr": 0.0003, "grad_norm": 0.15223429740935868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74763728380, "tokens_per_second": 150930.57722491908, "data_wait_s": 1.7693000700091943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510900, "loss": 0.5510307550430298, "lr": 0.0003, "grad_norm": 0.16456415781890485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74778295326, "tokens_per_second": 151104.70404062924, "data_wait_s": 1.678300031926483e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 511000, "loss": 0.5631042718887329, "lr": 0.0003, "grad_norm": 0.1702798864645056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74793099178, "tokens_per_second": 153885.9698552784, "data_wait_s": 1.3248998584458604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b489506d3d93b7f6fb13174670d8d320e8f3d17225fa5546a3f391156b37b19
-size 552642
+oid sha256:834905fba9cfc6a020812a73a40b4dc788828472500a4f42a9de18df09e00320
+size 554760