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": 76697.37666168743,
-      "required_pages_per_second_upper_estimate": 0.00530610247988783
+      "required_token_positions_per_second": 76872.71620637337,
+      "required_pages_per_second_upper_estimate": 0.005318232876432999
     },
     {
       "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": 9587.17208271093,
-      "required_pages_per_second_upper_estimate": 0.004013948748112146
+      "required_token_positions_per_second": 9609.089525796671,
+      "required_pages_per_second_upper_estimate": 0.004023125123843889
     },
     {
       "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": 19174.34416542186,
-      "required_pages_per_second_upper_estimate": 0.003993510228197471
+      "required_token_positions_per_second": 19218.179051593343,
+      "required_pages_per_second_upper_estimate": 0.004002639878983304
     },
     {
       "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": 28761.51624813279,
-      "required_pages_per_second_upper_estimate": 0.0017242089827830857
+      "required_token_positions_per_second": 28827.268577390012,
+      "required_pages_per_second_upper_estimate": 0.0017281507345240626
     },
     {
       "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": 28761.51624813279,
-      "required_pages_per_second_upper_estimate": 0.0022360813797521837
+      "required_token_positions_per_second": 28827.268577390012,
+      "required_pages_per_second_upper_estimate": 0.0022411933341380014
     },
     {
       "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": 28761.51624813279,
-      "required_pages_per_second_upper_estimate": 0.3689218487209347
+      "required_token_positions_per_second": 28827.268577390012,
+      "required_pages_per_second_upper_estimate": 0.36976524900129565
     }
   ],
   "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": 0.40830510079703686,
-    "producer_p95_seconds": 0.2960303850013588,
-    "gpu_update_seconds": 0.9612845081418446,
-    "producer_batches_per_second": 2.4491489282106396,
-    "gpu_batches_per_second": 1.0402747485580435,
-    "producer_headroom": 2.354328922821091,
+    "producer_batch_seconds": 0.49471321118755895,
+    "producer_p95_seconds": 0.324920243001543,
+    "gpu_update_seconds": 0.9590919072258222,
+    "producer_batches_per_second": 2.0213731458666735,
+    "gpu_batches_per_second": 1.0426529433373124,
+    "producer_headroom": 1.9386826256843277,
     "sustainable": true,
-    "buffer_seconds": 60.56092401293621,
+    "buffer_seconds": 60.4227901552268,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308766.30571486393,
-    "upload_bytes_per_second": 3189204.439867292,
+    "generation_bytes_per_second": 287190.917056129,
+    "upload_bytes_per_second": 3174727.580856075,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.25406149467887,
+    "estimated_full_shards_per_hour": 61.62448533785727,
     "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": 1.6119605509993562,
-      "bytes": 6041654,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.6828175019982154,
       "bytes": 10664698,
@@ -562,6 +556,12 @@
       "bytes": 5828306,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.8049468739991426,
+      "bytes": 4697774,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 830000, "loss": 0.5084052085876465, "lr": 3e-05, "grad_norm": 0.29305695782864816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121488869175, "tokens_per_second": 151431.7006384515, "data_wait_s": 7.247999747050926e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 830100, "loss": 0.5526085495948792, "lr": 3e-05, "grad_norm": 0.2950825940915703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121503568006, "tokens_per_second": 142074.65559539117, "data_wait_s": 9.081999451154843e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 830200, "loss": 0.568459153175354, "lr": 3e-05, "grad_norm": 0.34224297926513647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121518283207, "tokens_per_second": 149007.72947814903, "data_wait_s": 9.16499993763864e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 830300, "loss": 0.511318027973175, "lr": 3e-05, "grad_norm": 0.2898206888323544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121532931866, "tokens_per_second": 151441.41417938436, "data_wait_s": 1.4012999599799514e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3cfc4ac605cc35d7c50ba2a9a9e53d780219ce4653334aba24e7c3e8229dd93
-size 644499
+oid sha256:d70e95574ae48bf99594f32b0680c99129684e1748f728abbd0184b6c4acadfa
+size 646617