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": 74454.60210753139,
-      "required_pages_per_second_upper_estimate": 0.00645353620721699
+      "required_token_positions_per_second": 75687.99568780312,
+      "required_pages_per_second_upper_estimate": 0.006560443636747494
     },
     {
       "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": 9306.825263441424,
-      "required_pages_per_second_upper_estimate": 0.0037126317470246623
+      "required_token_positions_per_second": 9460.99946097539,
+      "required_pages_per_second_upper_estimate": 0.003774134139530633
     },
     {
       "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": 18613.650526882848,
-      "required_pages_per_second_upper_estimate": 0.004305940460333629
+      "required_token_positions_per_second": 18921.99892195078,
+      "required_pages_per_second_upper_estimate": 0.004377271434786189
     },
     {
       "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": 27920.475790324275,
-      "required_pages_per_second_upper_estimate": 0.0017002104882154205
+      "required_token_positions_per_second": 28382.998382926173,
+      "required_pages_per_second_upper_estimate": 0.0017283756874363755
     },
     {
       "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": 27920.475790324275,
-      "required_pages_per_second_upper_estimate": 0.002149395515162855
+      "required_token_positions_per_second": 28382.998382926173,
+      "required_pages_per_second_upper_estimate": 0.0021850017846858313
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27920.475790324275,
-      "required_pages_per_second_upper_estimate": 0.3466788654944221
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28382.998382926173,
+      "required_pages_per_second_upper_estimate": 0.35242184813099786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1051966699218454,
-    "producer_p95_seconds": 0.19027035199997044,
-    "gpu_update_seconds": 0.9902410047604311,
-    "producer_batches_per_second": 9.506004332104219,
-    "gpu_batches_per_second": 1.0098551718143907,
-    "producer_headroom": 9.413235281079894,
+    "producer_batch_seconds": 0.3162086590624824,
+    "producer_p95_seconds": 0.20367735699983314,
+    "gpu_update_seconds": 0.9741042728111379,
+    "producer_batches_per_second": 3.1624687412573396,
+    "gpu_batches_per_second": 1.0265841429009754,
+    "producer_headroom": 3.0805743134904353,
     "sustainable": true,
-    "buffer_seconds": 60.4047012903863,
+    "buffer_seconds": 60.394464914290545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334598.27331239753,
-    "upload_bytes_per_second": 4272348.418518677,
+    "generation_bytes_per_second": 309008.20145451126,
+    "upload_bytes_per_second": 4388834.867738892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.79700040368027,
+    "estimated_full_shards_per_hour": 66.30596668936256,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 811866,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.1758913479998228,
+      "bytes": 13372381,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -50,3 +50,4 @@
 {"step": 5000, "loss": 0.7549786865711212, "lr": 0.0003, "grad_norm": 0.16393758598977393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 731411255, "tokens_per_second": 147063.72191024804, "data_wait_s": 1.298300048802048e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5100, "loss": 0.7820186764001846, "lr": 0.0003, "grad_norm": 0.18854054384762203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 745997968, "tokens_per_second": 136184.03242160723, "data_wait_s": 1.439300012862077e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 5200, "loss": 0.734810758382082, "lr": 0.0003, "grad_norm": 0.18901925245297882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 760600430, "tokens_per_second": 148829.9078821076, "data_wait_s": 7.3940000220318325e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 5300, "loss": 0.7329708263278008, "lr": 0.0003, "grad_norm": 0.15539199647282062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 775327069, "tokens_per_second": 149697.5084817163, "data_wait_s": 9.624999620427843e-06, "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:c0c8b5c4e811f12c8ddcdf51f76c8a87013279628cd9f88b5c796bf8b0780332
-size 109215
+oid sha256:bc3f86ec4aad309f178447f5aa6716daa9e51c2f5b44461b68501f68fdf1754f
+size 111298