Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76053.13199712317,
-      "required_pages_per_second_upper_estimate": 0.006592092726609841
+      "required_token_positions_per_second": 75489.24839525507,
+      "required_pages_per_second_upper_estimate": 0.006543216725149837
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9506.641499640396,
-      "required_pages_per_second_upper_estimate": 0.0037923414311633943
+      "required_token_positions_per_second": 9436.156049406884,
+      "required_pages_per_second_upper_estimate": 0.0037642237312138518
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19013.282999280793,
-      "required_pages_per_second_upper_estimate": 0.004398388399531597
+      "required_token_positions_per_second": 18872.312098813767,
+      "required_pages_per_second_upper_estimate": 0.004365777262711662
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.92449892119,
-      "required_pages_per_second_upper_estimate": 0.001736713769504662
+      "required_token_positions_per_second": 28308.46814822065,
+      "required_pages_per_second_upper_estimate": 0.0017238371871727304
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.92449892119,
-      "required_pages_per_second_upper_estimate": 0.0021955427361308763
+      "required_token_positions_per_second": 28308.46814822065,
+      "required_pages_per_second_upper_estimate": 0.002179264240905306
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28519.92449892119,
-      "required_pages_per_second_upper_estimate": 0.35412201222942485
+      "required_token_positions_per_second": 28308.46814822065,
+      "required_pages_per_second_upper_estimate": 0.351496432052605
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07600292567192213,
-    "producer_p95_seconds": 0.20683527600067464,
-    "gpu_update_seconds": 0.9694275313052048,
-    "producer_batches_per_second": 13.157388234193087,
-    "gpu_batches_per_second": 1.0315366210547305,
-    "producer_headroom": 12.755134394297952,
+    "producer_batch_seconds": 0.3862888134999807,
+    "producer_p95_seconds": 0.19958992600004422,
+    "gpu_update_seconds": 0.9766688842093999,
+    "producer_batches_per_second": 2.5887366267210066,
+    "gpu_batches_per_second": 1.023888460222101,
+    "producer_headroom": 2.528338512731611,
     "sustainable": true,
-    "buffer_seconds": 60.1045069409227,
+    "buffer_seconds": 60.553470820982795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257961.09654810838,
-    "upload_bytes_per_second": 4700124.391248186,
+    "generation_bytes_per_second": 252762.77585305486,
+    "upload_bytes_per_second": 4753894.160515528,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.352446292232884,
+    "estimated_full_shards_per_hour": 54.237007681786864,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 865744,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.143092911000167,
+      "bytes": 10241728,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 7100, "loss": 0.7017961051315069, "lr": 0.0003, "grad_norm": 0.16912673231131298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1038636493, "tokens_per_second": 148824.46616029536, "data_wait_s": 1.4393999663298018e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 7200, "loss": 0.7542180214077234, "lr": 0.0003, "grad_norm": 0.16230978552127284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1053322124, "tokens_per_second": 150070.57520444802, "data_wait_s": 1.1235000783926807e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 7300, "loss": 0.7433256637305021, "lr": 0.0003, "grad_norm": 0.167053648832272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1067895954, "tokens_per_second": 148973.93984702454, "data_wait_s": 1.0682000720407814e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 7400, "loss": 0.7434984631836414, "lr": 0.0003, "grad_norm": 0.16470966525716607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1082563209, "tokens_per_second": 148560.87625178442, "data_wait_s": 1.1312999959045555e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a72b8ca04e9a194537019df76fd98b4b9d1e708b0a1309e1054ce917dc1acb66
-size 152958
+oid sha256:edc6200575d56a4f2f952d865411e64f38eb27fc61df83dd32f5540555108cee
+size 155041