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": 74721.93663688346,
-      "required_pages_per_second_upper_estimate": 0.006476708086668058
+      "required_token_positions_per_second": 73781.05112232354,
+      "required_pages_per_second_upper_estimate": 0.0063951545149185385
     },
     {
       "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": 9340.242079610432,
-      "required_pages_per_second_upper_estimate": 0.0037259622146204054
+      "required_token_positions_per_second": 9222.631390290442,
+      "required_pages_per_second_upper_estimate": 0.0036790455522141543
     },
     {
       "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": 18680.484159220865,
-      "required_pages_per_second_upper_estimate": 0.004321401244943274
+      "required_token_positions_per_second": 18445.262780580884,
+      "required_pages_per_second_upper_estimate": 0.0042669869187497884
     },
     {
       "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": 28020.7262388313,
-      "required_pages_per_second_upper_estimate": 0.0017063152145560443
+      "required_token_positions_per_second": 27667.89417087133,
+      "required_pages_per_second_upper_estimate": 0.001684829592784077
     },
     {
       "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": 28020.7262388313,
-      "required_pages_per_second_upper_estimate": 0.0021571130721999288
+      "required_token_positions_per_second": 27667.89417087133,
+      "required_pages_per_second_upper_estimate": 0.0021299510829066942
     },
     {
       "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": 28020.7262388313,
-      "required_pages_per_second_upper_estimate": 0.3479236405482114
+      "required_token_positions_per_second": 27667.89417087133,
+      "required_pages_per_second_upper_estimate": 0.34354264711711796
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23103490317186015,
-    "producer_p95_seconds": 0.18980877800004237,
-    "gpu_update_seconds": 0.9866981949127795,
-    "producer_batches_per_second": 4.328350332660036,
-    "gpu_batches_per_second": 1.0134811284299514,
-    "producer_headroom": 4.270775460185786,
+    "producer_batch_seconds": 0.3257991623827934,
+    "producer_p95_seconds": 0.16532143400036148,
+    "gpu_update_seconds": 0.9992809654848155,
+    "producer_batches_per_second": 3.0693756014788747,
+    "gpu_batches_per_second": 1.0007195518978345,
+    "producer_headroom": 3.067168614481346,
     "sustainable": true,
-    "buffer_seconds": 60.188589889679555,
+    "buffer_seconds": 60.95613889457375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460550.0537750822,
-    "upload_bytes_per_second": 5864368.293408163,
+    "generation_bytes_per_second": 413978.94844017766,
+    "upload_bytes_per_second": 6130657.500902683,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.82332048358298,
+    "estimated_full_shards_per_hour": 88.83024539855954,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 7480746,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.141071416000159,
+      "bytes": 10868715,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 44100, "loss": 0.6197555270045996, "lr": 0.0003, "grad_norm": 0.13371724287142478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6453322297, "tokens_per_second": 147163.8639494913, "data_wait_s": 1.3975999536341988e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 44200, "loss": 0.6590540930628777, "lr": 0.0003, "grad_norm": 0.12589594120792405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6468014811, "tokens_per_second": 149725.6945106827, "data_wait_s": 7.3430001066299155e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 44300, "loss": 0.6200189627707005, "lr": 0.0003, "grad_norm": 0.22031305739333554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6482772528, "tokens_per_second": 149493.1879755575, "data_wait_s": 1.0878000466618687e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 44400, "loss": 0.6678473874926567, "lr": 0.0003, "grad_norm": 0.14349906677865265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6497548429, "tokens_per_second": 151151.85950591083, "data_wait_s": 7.697000000916887e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:114517964737dfbd9481ed7e087c1d86a3403ed0b8a6af0b981a26dd50fe883d
-size 91239
+oid sha256:1682fb578590d36aafa811ed2b4007e0294daa1875e15c9bf6235a1f69aca28a
+size 93357