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": 76396.93655991535,
-      "required_pages_per_second_upper_estimate": 0.005285317336530157
+      "required_token_positions_per_second": 76426.13221591969,
+      "required_pages_per_second_upper_estimate": 0.005287337159755792
     },
     {
       "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": 9549.61706998942,
-      "required_pages_per_second_upper_estimate": 0.0039982252485234945
+      "required_token_positions_per_second": 9553.26652698996,
+      "required_pages_per_second_upper_estimate": 0.003999753199960293
     },
     {
       "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": 19099.23413997884,
-      "required_pages_per_second_upper_estimate": 0.003977866790682263
+      "required_token_positions_per_second": 19106.53305397992,
+      "required_pages_per_second_upper_estimate": 0.0039793869619833815
     },
     {
       "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": 28648.85120996826,
-      "required_pages_per_second_upper_estimate": 0.0017174548857746745
+      "required_token_positions_per_second": 28659.799580969884,
+      "required_pages_per_second_upper_estimate": 0.0017181112238990274
     },
     {
       "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": 28648.85120996826,
-      "required_pages_per_second_upper_estimate": 0.0022273221685960216
+      "required_token_positions_per_second": 28659.799580969884,
+      "required_pages_per_second_upper_estimate": 0.0022281733562846033
     },
     {
       "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": 28648.85120996826,
-      "required_pages_per_second_upper_estimate": 0.3674767025816531
+      "required_token_positions_per_second": 28659.799580969884,
+      "required_pages_per_second_upper_estimate": 0.3676171365294171
     }
   ],
   "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": 0.3075873138593721,
-    "producer_p95_seconds": 0.24570860499989067,
-    "gpu_update_seconds": 0.9650648745866635,
-    "producer_batches_per_second": 3.251109375912677,
-    "gpu_batches_per_second": 1.0361997688790603,
-    "producer_headroom": 3.1375314621326935,
+    "producer_batch_seconds": 0.0937192977265795,
+    "producer_p95_seconds": 0.2460688109986222,
+    "gpu_update_seconds": 0.9646962087745471,
+    "producer_batches_per_second": 10.670161047487154,
+    "gpu_batches_per_second": 1.0365957603070703,
+    "producer_headroom": 10.293463909524709,
     "sustainable": true,
-    "buffer_seconds": 60.799087098959795,
+    "buffer_seconds": 60.775861152796466,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483346.65032270737,
-    "upload_bytes_per_second": 5234078.703130761,
+    "generation_bytes_per_second": 489422.72928409866,
+    "upload_bytes_per_second": 5318654.918483729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.71493942509571,
+    "estimated_full_shards_per_hour": 105.01872452633114,
     "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": 1.3777619700003925,
-      "bytes": 819264,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.2043475289992784,
       "bytes": 13122126,
@@ -562,6 +556,12 @@
       "bytes": 6445004,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1019184790002328,
+      "bytes": 6210962,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -155,3 +155,4 @@
 {"step": 775500, "loss": 0.5328605771064758, "lr": 3e-05, "grad_norm": 0.24942127565431396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113511064431, "tokens_per_second": 153433.6788007891, "data_wait_s": 1.2311000318732113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775600, "loss": 0.5252447128295898, "lr": 3e-05, "grad_norm": 0.2516212729134852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113525843560, "tokens_per_second": 153074.31958093596, "data_wait_s": 1.1676998838083819e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775700, "loss": 0.5464670658111572, "lr": 3e-05, "grad_norm": 0.25042063006336235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113540419679, "tokens_per_second": 150802.97514128106, "data_wait_s": 1.2832999345846474e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 775800, "loss": 0.5179250240325928, "lr": 3e-05, "grad_norm": 0.2545529767185388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113554958333, "tokens_per_second": 150953.8647159574, "data_wait_s": 1.4745999578735791e-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 sha256:1a0ece42f95f54b5a6fddbd39455ffe9d5014bcee779353e8241687c19b9a55d
-size 335040
+oid sha256:fa9b58097c167e23eacbe8c9967dbe548da93f94fd06e7ef5ff1ee33e7deffff
+size 337158