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": 77367.0652507272,
-      "required_pages_per_second_upper_estimate": 0.005352433090369212
+      "required_token_positions_per_second": 77074.21938440595,
+      "required_pages_per_second_upper_estimate": 0.005332173333841082
     },
     {
       "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": 9670.8831563409,
-      "required_pages_per_second_upper_estimate": 0.004048996826555016
+      "required_token_positions_per_second": 9634.277423050744,
+      "required_pages_per_second_upper_estimate": 0.004033670770441063
     },
     {
       "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": 19341.7663126818,
-      "required_pages_per_second_upper_estimate": 0.004028379846252783
+      "required_token_positions_per_second": 19268.554846101488,
+      "required_pages_per_second_upper_estimate": 0.004013131828480313
     },
     {
       "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": 29012.6494690227,
-      "required_pages_per_second_upper_estimate": 0.0017392640359172105
+      "required_token_positions_per_second": 28902.83226915223,
+      "required_pages_per_second_upper_estimate": 0.001732680662466131
     },
     {
       "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": 29012.6494690227,
-      "required_pages_per_second_upper_estimate": 0.002255605883058075
+      "required_token_positions_per_second": 28902.83226915223,
+      "required_pages_per_second_upper_estimate": 0.0022470680788029595
     },
     {
       "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": 29012.6494690227,
-      "required_pages_per_second_upper_estimate": 0.37214311603266637
+      "required_token_positions_per_second": 28902.83226915223,
+      "required_pages_per_second_upper_estimate": 0.37073449890525045
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4273295151873526,
-    "producer_p95_seconds": 0.32962078600394307,
-    "gpu_update_seconds": 0.9529636384819056,
-    "producer_batches_per_second": 2.340114512243727,
-    "gpu_batches_per_second": 1.0493579813737957,
-    "producer_headroom": 2.230044040052092,
+    "producer_batch_seconds": 0.10960970971069628,
+    "producer_p95_seconds": 0.3167601340028341,
+    "gpu_update_seconds": 0.9565844531267096,
+    "producer_batches_per_second": 9.123279339388807,
+    "gpu_batches_per_second": 1.0453860051053323,
+    "producer_headroom": 8.72718717759145,
     "sustainable": true,
-    "buffer_seconds": 60.03670922436005,
+    "buffer_seconds": 60.264820546982705,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319420.53383669775,
-    "upload_bytes_per_second": 4748345.550763695,
+    "generation_bytes_per_second": 319661.7300228058,
+    "upload_bytes_per_second": 4726793.098938063,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.54021083188724,
+    "estimated_full_shards_per_hour": 68.59196591866618,
     "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": 1.2705231469990395,
-      "bytes": 7122874,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2547058620002645,
       "bytes": 6059042,
@@ -562,6 +556,12 @@
       "bytes": 6495633,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.523345616995357,
+      "bytes": 6337115,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 798700, "loss": 0.5392810106277466, "lr": 3e-05, "grad_norm": 0.2633574970401355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116906656287, "tokens_per_second": 151940.5872339052, "data_wait_s": 1.1012001777999103e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 798800, "loss": 0.5650307536125183, "lr": 3e-05, "grad_norm": 0.2704147929740986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116921270196, "tokens_per_second": 151874.6052369894, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 798900, "loss": 0.4903576970100403, "lr": 3e-05, "grad_norm": 0.2642801376038097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116935902606, "tokens_per_second": 152388.29585401132, "data_wait_s": 9.620001947041601e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 799000, "loss": 0.49883708357810974, "lr": 3e-05, "grad_norm": 0.27126486284628265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116950482357, "tokens_per_second": 151958.34753286018, "data_wait_s": 1.377000444335863e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0adc8137478988ffb5f5bc91d350419ec2e7eb31e28502524e6fd958e2c019a1
-size 829548
+oid sha256:2b6b4444eff6fd65ec7356c80de2d7c9fccadad6ef59ee712c198e2dc140fceb
+size 831666