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": 75396.34596431271,
-      "required_pages_per_second_upper_estimate": 0.006535164177894812
+      "required_token_positions_per_second": 74575.6696782461,
+      "required_pages_per_second_upper_estimate": 0.006464030037403588
     },
     {
       "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": 9424.54324553909,
-      "required_pages_per_second_upper_estimate": 0.0037595912101240983
+      "required_token_positions_per_second": 9321.958709780762,
+      "required_pages_per_second_upper_estimate": 0.0037186687050346106
     },
     {
       "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": 18849.08649107818,
-      "required_pages_per_second_upper_estimate": 0.004360404427118768
+      "required_token_positions_per_second": 18643.917419561523,
+      "required_pages_per_second_upper_estimate": 0.004312942173275723
     },
     {
       "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": 28273.629736617266,
-      "required_pages_per_second_upper_estimate": 0.0017217157106891037
+      "required_token_positions_per_second": 27965.876129342287,
+      "required_pages_per_second_upper_estimate": 0.0017029751306644474
     },
     {
       "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": 28273.629736617266,
-      "required_pages_per_second_upper_estimate": 0.0021765822835411817
+      "required_token_positions_per_second": 27965.876129342287,
+      "required_pages_per_second_upper_estimate": 0.0021528905589366437
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28273.629736617266,
-      "required_pages_per_second_upper_estimate": 0.3510638555771542
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27965.876129342287,
+      "required_pages_per_second_upper_estimate": 0.3472425857598655
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.38331178391422327,
-    "producer_p95_seconds": 0.3835994449982536,
-    "gpu_update_seconds": 0.9778723233470441,
-    "producer_batches_per_second": 2.608842310529587,
-    "gpu_batches_per_second": 1.022628390357974,
-    "producer_headroom": 2.551114691443638,
+    "producer_batch_seconds": 0.14102911641430182,
+    "producer_p95_seconds": 0.3827710149998893,
+    "gpu_update_seconds": 0.9886334285444122,
+    "producer_batches_per_second": 7.090734349226835,
+    "gpu_batches_per_second": 1.0114972558355861,
+    "producer_headroom": 7.010137010573757,
     "sustainable": true,
-    "buffer_seconds": 60.62808404751674,
+    "buffer_seconds": 60.30663914120914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390636.919700897,
-    "upload_bytes_per_second": 6319780.77917444,
+    "generation_bytes_per_second": 463921.03633565013,
+    "upload_bytes_per_second": 6166340.979345969,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.82158940572913,
+    "estimated_full_shards_per_hour": 99.54665486862305,
     "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": [
-    {
-      "seconds": 1.1346876719981083,
-      "bytes": 10084333,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9850222569984908,
       "bytes": 2356425,
@@ -562,6 +556,12 @@
       "bytes": 5648372,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 2.317049978002615,
+      "bytes": 7094864,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 32100, "loss": 0.6099816374480724, "lr": 0.0003, "grad_norm": 0.14531868021638142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4697595341, "tokens_per_second": 150225.49777577218, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32200, "loss": 0.6732552777975798, "lr": 0.0003, "grad_norm": 0.14508445797455272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4712325524, "tokens_per_second": 150656.4921985203, "data_wait_s": 9.315997886005789e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32300, "loss": 0.6117614042013884, "lr": 0.0003, "grad_norm": 0.13833622898780742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4727005304, "tokens_per_second": 149149.09986046032, "data_wait_s": 1.377000444335863e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 32400, "loss": 0.6780863795429468, "lr": 0.0003, "grad_norm": 0.12556539759495675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4741489249, "tokens_per_second": 147126.84428959162, "data_wait_s": 9.2410045908764e-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:dc019f67ef8e209472fed2407c8aac1a7cbb16a86f2b236080e062f690dd8735
-size 683203
+oid sha256:389bb6026b59b1be24c2e4a010b75c4ba182b90dc021cea4f3e5390ea2e437c3
+size 685321