Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76965.04228639213,
-      "required_pages_per_second_upper_estimate": 0.0053246202088747165
+      "required_token_positions_per_second": 76315.31528464453,
+      "required_pages_per_second_upper_estimate": 0.005279670587319471
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9620.630285799016,
-      "required_pages_per_second_upper_estimate": 0.004027956999058398
+      "required_token_positions_per_second": 9539.414410580566,
+      "required_pages_per_second_upper_estimate": 0.003993953608084764
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19241.26057159803,
-      "required_pages_per_second_upper_estimate": 0.004007447150899665
+      "required_token_positions_per_second": 19078.828821161133,
+      "required_pages_per_second_upper_estimate": 0.003973616900896979
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28861.89085739705,
-      "required_pages_per_second_upper_estimate": 0.001730226287345324
+      "required_token_positions_per_second": 28618.243231741697,
+      "required_pages_per_second_upper_estimate": 0.0017156199842156689
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28861.89085739705,
-      "required_pages_per_second_upper_estimate": 0.002243885064121233
+      "required_token_positions_per_second": 28618.243231741697,
+      "required_pages_per_second_upper_estimate": 0.0022249425329191747
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28861.89085739705,
-      "required_pages_per_second_upper_estimate": 0.37020934643471803
+      "required_token_positions_per_second": 28618.243231741697,
+      "required_pages_per_second_upper_estimate": 0.36708409630124933
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11206410968759428,
-    "producer_p95_seconds": 0.24095538199981092,
-    "gpu_update_seconds": 0.95794139533703,
-    "producer_batches_per_second": 8.92346356730751,
-    "gpu_batches_per_second": 1.0439051959417336,
-    "producer_headroom": 8.548155140905706,
+    "producer_batch_seconds": 0.4152253710546603,
+    "producer_p95_seconds": 0.25926168600017263,
+    "gpu_update_seconds": 0.9660970373378629,
+    "producer_batches_per_second": 2.4083306794573494,
+    "gpu_batches_per_second": 1.0350927094813982,
+    "producer_headroom": 2.3266811343536276,
     "sustainable": true,
-    "buffer_seconds": 60.35030790623289,
+    "buffer_seconds": 60.864113352285365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507144.12652508303,
-    "upload_bytes_per_second": 5557020.233643524,
+    "generation_bytes_per_second": 457932.35297836777,
+    "upload_bytes_per_second": 5667296.694447435,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.821323841232,
+    "estimated_full_shards_per_hour": 98.26162282956385,
     "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": 1.0184807959994941,
-      "bytes": 301114,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 1.0933929989987519,
       "bytes": 5793673,
@@ -562,6 +556,12 @@
       "bytes": 5765223,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9962533819998498,
+      "bytes": 8524201,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -156,3 +156,4 @@
 {"step": 412700, "loss": 0.6010996699333191, "lr": 0.0003, "grad_norm": 0.15243001164073552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60408700050, "tokens_per_second": 151830.57496950688, "data_wait_s": 1.456500103813596e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412800, "loss": 0.5927441716194153, "lr": 0.0003, "grad_norm": 0.17216612791240354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60423269331, "tokens_per_second": 151040.5523618154, "data_wait_s": 1.3699000191991217e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412900, "loss": 0.6292275190353394, "lr": 0.0003, "grad_norm": 0.16825339119515065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60437878206, "tokens_per_second": 150782.78168671072, "data_wait_s": 9.114999556913972e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 413000, "loss": 0.5962061882019043, "lr": 0.0003, "grad_norm": 0.1740489759002041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60452450337, "tokens_per_second": 151370.53684027633, "data_wait_s": 1.420499938831199e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f471b36ed7305132fa6b7f31f0a4bfd259b01aef78c2feb0650e8950adcbd8e
-size 337158
+oid sha256:389660a5180d6fbab6e37bc6a6948938548b268c717afe05d36b38464e5affc9
+size 339276