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": 76258.62531268995,
-      "required_pages_per_second_upper_estimate": 0.005275748643520796
+      "required_token_positions_per_second": 76366.33618496417,
+      "required_pages_per_second_upper_estimate": 0.005283200331588382
     },
     {
       "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": 9532.328164086244,
-      "required_pages_per_second_upper_estimate": 0.003990986744655245
+      "required_token_positions_per_second": 9545.79202312052,
+      "required_pages_per_second_upper_estimate": 0.003996623781275548
     },
     {
       "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": 19064.65632817249,
-      "required_pages_per_second_upper_estimate": 0.003970665144361218
+      "required_token_positions_per_second": 19091.58404624104,
+      "required_pages_per_second_upper_estimate": 0.003976273477903218
     },
     {
       "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": 28596.984492258736,
-      "required_pages_per_second_upper_estimate": 0.001714345555243881
+      "required_token_positions_per_second": 28637.376069361562,
+      "required_pages_per_second_upper_estimate": 0.0017167669686168282
     },
     {
       "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": 28596.984492258736,
-      "required_pages_per_second_upper_estimate": 0.002223289759431687
+      "required_token_positions_per_second": 28637.376069361562,
+      "required_pages_per_second_upper_estimate": 0.0022264300268876602
     },
     {
       "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": 28596.984492258736,
-      "required_pages_per_second_upper_estimate": 0.36681141201701795
+      "required_token_positions_per_second": 28637.376069361562,
+      "required_pages_per_second_upper_estimate": 0.3673295117989965
     }
   ],
   "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": 0.10205526479691684,
-    "producer_p95_seconds": 0.2866812979991664,
-    "gpu_update_seconds": 0.9668152251327189,
-    "producer_batches_per_second": 9.798612565357928,
-    "gpu_batches_per_second": 1.0343238025267194,
-    "producer_headroom": 9.473447813364814,
+    "producer_batch_seconds": 0.4998368583280808,
+    "producer_p95_seconds": 0.2842813349998323,
+    "gpu_update_seconds": 0.9654515809351657,
+    "producer_batches_per_second": 2.000652779678813,
+    "gpu_batches_per_second": 1.0357847247309593,
+    "producer_headroom": 1.931533389043244,
     "sustainable": true,
-    "buffer_seconds": 60.90935918336129,
+    "buffer_seconds": 60.82344959891544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364236.79064211866,
-    "upload_bytes_per_second": 4870983.946111981,
+    "generation_bytes_per_second": 307269.6231738124,
+    "upload_bytes_per_second": 4736748.722279621,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.1567362732665,
+    "estimated_full_shards_per_hour": 65.93290826235561,
     "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": 1.2755690529993444,
-      "bytes": 16950690,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.0866290939993632,
       "bytes": 1216935,
@@ -562,6 +556,12 @@
       "bytes": 2637387,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4529079339990858,
+      "bytes": 5933529,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -306,3 +306,4 @@
 {"step": 790600, "loss": 0.59759122133255, "lr": 3e-05, "grad_norm": 0.3038910553041401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115721224740, "tokens_per_second": 150316.62702318624, "data_wait_s": 1.3362001482164487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790700, "loss": 0.5069711804389954, "lr": 3e-05, "grad_norm": 0.2505566734304559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115735872939, "tokens_per_second": 152687.89437593988, "data_wait_s": 9.497998689766973e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 790800, "loss": 0.534726083278656, "lr": 3e-05, "grad_norm": 0.2632059355800747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115750515759, "tokens_per_second": 151996.63088258673, "data_wait_s": 1.0162999387830496e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 790900, "loss": 0.5075185894966125, "lr": 3e-05, "grad_norm": 0.25775752807402286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115765132292, "tokens_per_second": 151411.71402798707, "data_wait_s": 8.085000445134938e-06, "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 sha256:70e00650b2506bfb855205ab0ec78ca2e4ced5c748a6293a1bdea0805908777e
-size 657207
+oid sha256:d7b0f6feb0ae9e9faf61a7bccdb26f1e6883873dad33651de8f032f84a95d556
+size 659325