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": 74759.73872587517,
-      "required_pages_per_second_upper_estimate": 0.00647998467593334
+      "required_token_positions_per_second": 75362.79979938698,
+      "required_pages_per_second_upper_estimate": 0.006532256481341027
     },
     {
       "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": 9344.967340734396,
-      "required_pages_per_second_upper_estimate": 0.0037278471919317043
+      "required_token_positions_per_second": 9420.349974923372,
+      "required_pages_per_second_upper_estimate": 0.003757918451780506
     },
     {
       "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": 18689.934681468792,
-      "required_pages_per_second_upper_estimate": 0.004323587456941821
+      "required_token_positions_per_second": 18840.699949846745,
+      "required_pages_per_second_upper_estimate": 0.004358464348402961
     },
     {
       "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": 28034.902022203187,
-      "required_pages_per_second_upper_estimate": 0.0017071784453888048
+      "required_token_positions_per_second": 28261.04992477012,
+      "required_pages_per_second_upper_estimate": 0.0017209496661487844
     },
     {
       "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": 28034.902022203187,
-      "required_pages_per_second_upper_estimate": 0.0021582043632449766
+      "required_token_positions_per_second": 28261.04992477012,
+      "required_pages_per_second_upper_estimate": 0.00217561385480204
     },
     {
       "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": 28034.902022203187,
-      "required_pages_per_second_upper_estimate": 0.3480996563343952
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28261.04992477012,
+      "required_pages_per_second_upper_estimate": 0.3509076564159345
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3393074827889393,
-    "producer_p95_seconds": 0.2730642649985384,
-    "gpu_update_seconds": 0.9861992732524348,
-    "producer_batches_per_second": 2.947179330619224,
-    "gpu_batches_per_second": 1.0139938520762148,
-    "producer_headroom": 2.9065061140012762,
+    "producer_batch_seconds": 0.07716415895299633,
+    "producer_p95_seconds": 0.21215325299999677,
+    "gpu_update_seconds": 0.9783076026403111,
+    "producer_batches_per_second": 12.95938442884006,
+    "gpu_batches_per_second": 1.022173391376234,
+    "producer_headroom": 12.678264312272697,
     "sustainable": true,
-    "buffer_seconds": 60.15815566839853,
+    "buffer_seconds": 60.65507136369929,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339331.70434015983,
-    "upload_bytes_per_second": 6243179.083831291,
+    "generation_bytes_per_second": 352918.2639325382,
+    "upload_bytes_per_second": 6368556.419869212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.81268451360317,
+    "estimated_full_shards_per_hour": 75.72804392320738,
     "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": 1.0381730779990903,
-      "bytes": 7573812,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.756838974000857,
       "bytes": 1060731,
@@ -562,6 +556,12 @@
       "bytes": 772407,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.406286874000216,
+      "bytes": 18261180,
+      "pages": 44,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -170,3 +170,4 @@
 {"step": 57000, "loss": 0.6150839980691671, "lr": 0.0003, "grad_norm": 0.13519484046330424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8342772660, "tokens_per_second": 148430.38882894648, "data_wait_s": 1.074399915523827e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57100, "loss": 0.6742852330207825, "lr": 0.0003, "grad_norm": 0.15001693965512816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8357399409, "tokens_per_second": 149993.12242864975, "data_wait_s": 1.1885000276379287e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 57200, "loss": 0.6306101009249687, "lr": 0.0003, "grad_norm": 0.13996838550861912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8371995313, "tokens_per_second": 148911.11273361236, "data_wait_s": 1.0081999789690599e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 57300, "loss": 0.600916426628828, "lr": 0.0003, "grad_norm": 0.13766784713913385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8386571728, "tokens_per_second": 149561.33725087225, "data_wait_s": 1.091299782274291e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2b5bef4816812b34b80d0ef872068806d52d39951da375a3626e278f153b88e
-size 366810
+oid sha256:5b9aaba9e6c492f127babf5d575f3aac0dcebf7c9b024ef7c1a01a128b08cc84
+size 368928