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": 75314.37843676031,
-      "required_pages_per_second_upper_estimate": 0.006528059440351368
+      "required_token_positions_per_second": 75199.73225111033,
+      "required_pages_per_second_upper_estimate": 0.0065181221995473295
     },
     {
       "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": 9414.297304595038,
-      "required_pages_per_second_upper_estimate": 0.003755503951091048
+      "required_token_positions_per_second": 9399.96653138879,
+      "required_pages_per_second_upper_estimate": 0.003749787191394922
     },
     {
       "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": 18828.594609190077,
-      "required_pages_per_second_upper_estimate": 0.00435566399088876
+      "required_token_positions_per_second": 18799.93306277758,
+      "required_pages_per_second_upper_estimate": 0.0043490336465522176
     },
     {
       "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": 28242.891913785115,
-      "required_pages_per_second_upper_estimate": 0.0017198439385475185
+      "required_token_positions_per_second": 28199.899594166374,
+      "required_pages_per_second_upper_estimate": 0.001717225931846006
     },
     {
       "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": 28242.891913785115,
-      "required_pages_per_second_upper_estimate": 0.002174216000851823
+      "required_token_positions_per_second": 28199.899594166374,
+      "required_pages_per_second_upper_estimate": 0.002170906333077215
     },
     {
       "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": 28242.891913785115,
-      "required_pages_per_second_upper_estimate": 0.35068219469045425
+      "required_token_positions_per_second": 28199.899594166374,
+      "required_pages_per_second_upper_estimate": 0.35014837396682735
     }
   ],
   "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": 0.07521616482033266,
-    "producer_p95_seconds": 0.18666884699996444,
-    "gpu_update_seconds": 0.9789365793134395,
-    "producer_batches_per_second": 13.295014474464097,
-    "gpu_batches_per_second": 1.021516634613177,
-    "producer_headroom": 13.014975991554548,
+    "producer_batch_seconds": 0.3529192845078484,
+    "producer_p95_seconds": 0.18607838200023252,
+    "gpu_update_seconds": 0.980429022723168,
+    "producer_batches_per_second": 2.833509088046906,
+    "gpu_batches_per_second": 1.0199616462010408,
+    "producer_headroom": 2.778054546071043,
     "sustainable": true,
-    "buffer_seconds": 60.694067917433244,
+    "buffer_seconds": 60.786599408836416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357989.3202225032,
-    "upload_bytes_per_second": 4166339.6981900632,
+    "generation_bytes_per_second": 323402.4041736195,
+    "upload_bytes_per_second": 4268388.328580749,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.81617455488512,
+    "estimated_full_shards_per_hour": 69.39462751299324,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 900296,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.093376490999617,
+      "bytes": 11563576,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -46,3 +46,4 @@
 {"step": 4600, "loss": 0.7736689057201147, "lr": 0.0003, "grad_norm": 0.17025524734238284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 672773231, "tokens_per_second": 149800.9268492934, "data_wait_s": 1.0876000487769488e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4700, "loss": 0.803300105035305, "lr": 0.0003, "grad_norm": 0.18988430892715008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 687479992, "tokens_per_second": 149042.21413900176, "data_wait_s": 1.7197000488522463e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 4800, "loss": 0.8045641928911209, "lr": 0.0003, "grad_norm": 0.18913845286306885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 702310909, "tokens_per_second": 150303.99248884092, "data_wait_s": 1.3591999959317036e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 4900, "loss": 0.7926367856562138, "lr": 0.0003, "grad_norm": 0.18977786811464437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 716901524, "tokens_per_second": 148439.51503401782, "data_wait_s": 1.1121999705210328e-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:d19244854ef7e36fa545263a450fe39bf0380339027ea6a4817cf468b8ad21e4
-size 100110
+oid sha256:330d506ab9cd2c805390e8e9c859be0b41cd717b017ece02611d0883b92d4d40
+size 102193