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": 73974.25600705274,
-      "required_pages_per_second_upper_estimate": 0.006411901024653562
+      "required_token_positions_per_second": 74549.3016000283,
+      "required_pages_per_second_upper_estimate": 0.006461744519212952
     },
     {
       "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": 9246.782000881592,
-      "required_pages_per_second_upper_estimate": 0.0036886795918627704
+      "required_token_positions_per_second": 9318.662700003537,
+      "required_pages_per_second_upper_estimate": 0.003717353877454738
     },
     {
       "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": 18493.564001763185,
-      "required_pages_per_second_upper_estimate": 0.004278160556197854
+      "required_token_positions_per_second": 18637.325400007074,
+      "required_pages_per_second_upper_estimate": 0.004311417225567383
     },
     {
       "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": 27740.346002644776,
-      "required_pages_per_second_upper_estimate": 0.0016892415292136963
+      "required_token_positions_per_second": 27955.98810001061,
+      "required_pages_per_second_upper_estimate": 0.0017023730015566288
     },
     {
       "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": 27740.346002644776,
-      "required_pages_per_second_upper_estimate": 0.0021355286254761212
+      "required_token_positions_per_second": 27955.98810001061,
+      "required_pages_per_second_upper_estimate": 0.0021521293510668744
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27740.346002644776,
-      "required_pages_per_second_upper_estimate": 0.34444225638706155
+      "required_token_positions_per_second": 27955.98810001061,
+      "required_pages_per_second_upper_estimate": 0.34711980952867144
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.19042130631250487,
-    "producer_p95_seconds": 0.16449861099999907,
-    "gpu_update_seconds": 0.9966710580092991,
-    "producer_batches_per_second": 5.251513180772306,
-    "gpu_batches_per_second": 1.0033400608595477,
-    "producer_headroom": 5.234031198030114,
+    "producer_batch_seconds": 0.06532604078904214,
+    "producer_p95_seconds": 0.15224569000019983,
+    "gpu_update_seconds": 0.9889831080586813,
+    "producer_batches_per_second": 15.307831117904533,
+    "gpu_batches_per_second": 1.0111396158858004,
+    "producer_headroom": 15.139186396622621,
     "sustainable": true,
-    "buffer_seconds": 60.79693453856724,
+    "buffer_seconds": 60.32796959157956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535834.696413683,
-    "upload_bytes_per_second": 2996380.5200971365,
+    "generation_bytes_per_second": 530822.0331597126,
+    "upload_bytes_per_second": 3118117.5216924865,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.97765225704067,
+    "estimated_full_shards_per_hour": 113.90205141156704,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 7503017,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8951566169998841,
+      "bytes": 8341517,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -21,3 +21,4 @@
 {"step": 2100, "loss": 0.8481050059199333, "lr": 0.0003, "grad_norm": 0.23960433516464869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 307689242, "tokens_per_second": 147659.70686422856, "data_wait_s": 1.485599977968377e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 2200, "loss": 0.8594202846288681, "lr": 0.0003, "grad_norm": 0.21901127440206275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 322336648, "tokens_per_second": 149103.51465016266, "data_wait_s": 1.2503000107244588e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 2300, "loss": 0.8143489807844162, "lr": 0.0003, "grad_norm": 0.20700560284367647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 336933125, "tokens_per_second": 147882.52259958148, "data_wait_s": 9.175999821309233e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 2400, "loss": 0.873838447034359, "lr": 0.0003, "grad_norm": 0.22909544058462067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 351510809, "tokens_per_second": 148535.49089934255, "data_wait_s": 1.1372000244591618e-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:c9e8f3412e8281061261564f281cccea59a87a0b645dc4c55f484b74457758e1
-size 48035
+oid sha256:1a3ddc0542b23614d78a2442f0e6d0b7d1959135798011396e6a254cef04db60
+size 50118