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": 77763.58674823606,
-      "required_pages_per_second_upper_estimate": 0.005379865367623508
+      "required_token_positions_per_second": 77329.78701120264,
+      "required_pages_per_second_upper_estimate": 0.005349854095261471
     },
     {
       "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": 9720.448343529508,
-      "required_pages_per_second_upper_estimate": 0.0040697487353920795
+      "required_token_positions_per_second": 9666.22337640033,
+      "required_pages_per_second_upper_estimate": 0.00404704587400117
     },
     {
       "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": 19440.896687059016,
-      "required_pages_per_second_upper_estimate": 0.004049026088991784
+      "required_token_positions_per_second": 19332.44675280066,
+      "required_pages_per_second_upper_estimate": 0.004026438827702863
     },
     {
       "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": 29161.345030588524,
-      "required_pages_per_second_upper_estimate": 0.0017481781077881055
+      "required_token_positions_per_second": 28998.670129200993,
+      "required_pages_per_second_upper_estimate": 0.001738425995840115
     },
     {
       "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": 29161.345030588524,
-      "required_pages_per_second_upper_estimate": 0.002267166308927164
+      "required_token_positions_per_second": 28998.670129200993,
+      "required_pages_per_second_upper_estimate": 0.002254519050875555
     },
     {
       "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": 29161.345030588524,
-      "required_pages_per_second_upper_estimate": 0.3740504230395778
+      "required_token_positions_per_second": 28998.670129200993,
+      "required_pages_per_second_upper_estimate": 0.37196380407127916
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 1.5911978684999895,
-    "producer_p95_seconds": 3.0803219919999947,
-    "gpu_update_seconds": 0.9481044159999783,
-    "producer_batches_per_second": 0.6284573526626783,
-    "gpu_batches_per_second": 1.0547361483864484,
-    "producer_headroom": 0.5958431913271411,
-    "sustainable": false,
-    "buffer_seconds": 60.67868262399861,
-    "buffer_drain_seconds": 150.13648495308266,
+    "producer_batch_seconds": 0.052906279765622344,
+    "producer_p95_seconds": 0.14222698599996875,
+    "gpu_update_seconds": 0.9534230320499801,
+    "producer_batches_per_second": 18.90134790104414,
+    "gpu_batches_per_second": 1.048852362890661,
+    "producer_headroom": 18.020980425645032,
+    "sustainable": true,
+    "buffer_seconds": 60.065651019148746,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1711,
+    "upload_pages_per_commit": 18,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

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

@@ -0,0 +1 @@
+{"step": 760100, "loss": 0.5428827404975891, "lr": 3e-05, "grad_norm": 0.23948048064094665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111256501409, "tokens_per_second": 152889.36247226436, "data_wait_s": 1.5780000012455275e-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:bfbbc9eb4feafdffac6c531a1283b82aa0ca75d088f7cc2d67b6632c46205d19
-size 165
+oid sha256:be162eb88c389e170bd0556c97653a0a3c6d227cc4eedbfb6754c4cfad5d1ff5
+size 2283