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": 76608.11503721995,
-      "required_pages_per_second_upper_estimate": 0.005299927153591666
+      "required_token_positions_per_second": 74730.32238159912,
+      "required_pages_per_second_upper_estimate": 0.005170017100596567
     },
     {
       "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": 9576.014379652494,
-      "required_pages_per_second_upper_estimate": 0.004009277250840914
+      "required_token_positions_per_second": 9341.29029769989,
+      "required_pages_per_second_upper_estimate": 0.00391100317932357
     },
     {
       "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": 19152.028759304987,
-      "required_pages_per_second_upper_estimate": 0.00398886251760016
+      "required_token_positions_per_second": 18682.58059539978,
+      "required_pages_per_second_upper_estimate": 0.003891088845239319
     },
     {
       "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": 28728.04313895748,
-      "required_pages_per_second_upper_estimate": 0.0017222023209985042
+      "required_token_positions_per_second": 28023.87089309967,
+      "required_pages_per_second_upper_estimate": 0.0016799882700680913
     },
     {
       "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": 28728.04313895748,
-      "required_pages_per_second_upper_estimate": 0.0022334789927464507
+      "required_token_positions_per_second": 28023.87089309967,
+      "required_pages_per_second_upper_estimate": 0.0021787326979573795
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28728.04313895748,
-      "required_pages_per_second_upper_estimate": 0.36849249161705827
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28023.87089309967,
+      "required_pages_per_second_upper_estimate": 0.3594601261284446
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11191009941398988,
-    "producer_p95_seconds": 0.3243526459991699,
-    "gpu_update_seconds": 0.9624045698576366,
-    "producer_batches_per_second": 8.935744005558359,
-    "gpu_batches_per_second": 1.0390640602921541,
-    "producer_headroom": 8.599800866027346,
+    "producer_batch_seconds": 0.3776037906407623,
+    "producer_p95_seconds": 0.3380591800014372,
+    "gpu_update_seconds": 0.9865874741382633,
+    "producer_batches_per_second": 2.6482784992785238,
+    "gpu_batches_per_second": 1.01359486737195,
+    "producer_headroom": 2.6127583954178695,
     "sustainable": true,
-    "buffer_seconds": 60.6314879010311,
+    "buffer_seconds": 60.18183592243406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338247.8083685125,
-    "upload_bytes_per_second": 4786828.827312803,
+    "generation_bytes_per_second": 319743.01194772904,
+    "upload_bytes_per_second": 4846749.022576793,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.58010566989451,
+    "estimated_full_shards_per_hour": 68.60940712760834,
     "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": 1.0274238659985713,
-      "bytes": 6279174,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.6521351509982196,
       "bytes": 5784763,
@@ -562,6 +556,12 @@
       "bytes": 625562,
       "pages": 19,
       "shards": 1
+    },
+    {
+      "seconds": 1.4279748699991615,
+      "bytes": 13559996,
+      "pages": 51,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 795600, "loss": 0.5383407473564148, "lr": 3e-05, "grad_norm": 0.2748085125557036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116453264956, "tokens_per_second": 154613.32209299205, "data_wait_s": 9.90999978967011e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 795700, "loss": 0.6049256920814514, "lr": 3e-05, "grad_norm": 0.28892878948135853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116468097835, "tokens_per_second": 126994.52916165971, "data_wait_s": 9.45699866861105e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 795800, "loss": 0.5575957894325256, "lr": 3e-05, "grad_norm": 0.27267736367051293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116482759554, "tokens_per_second": 151456.69218906824, "data_wait_s": 8.636001439299434e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 795900, "loss": 0.5025835633277893, "lr": 3e-05, "grad_norm": 0.2636277595490215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116497429120, "tokens_per_second": 152916.09176264008, "data_wait_s": 8.536000677850097e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e52bbe18786a68e199b436a88dda072c21e9b893ae66c9cee6d09d2359850a7a
-size 763890
+oid sha256:c3b6b7c423bad87d61d3e8d653f004838b8f16d1f7c0da3a70ce5bfca7132ab8
+size 766008