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": 75209.74369519857,
-      "required_pages_per_second_upper_estimate": 0.006518989966147125
+      "required_token_positions_per_second": 74589.98429269603,
+      "required_pages_per_second_upper_estimate": 0.00646527079190405
     },
     {
       "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": 9401.217961899822,
-      "required_pages_per_second_upper_estimate": 0.003750286405736326
+      "required_token_positions_per_second": 9323.748036587003,
+      "required_pages_per_second_upper_estimate": 0.0037193824942504403
     },
     {
       "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": 18802.435923799643,
-      "required_pages_per_second_upper_estimate": 0.004349612639400822
+      "required_token_positions_per_second": 18647.496073174007,
+      "required_pages_per_second_upper_estimate": 0.004313770031806818
     },
     {
       "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": 28203.653885699463,
-      "required_pages_per_second_upper_estimate": 0.0017174545485031255
+      "required_token_positions_per_second": 27971.24410976101,
+      "required_pages_per_second_upper_estimate": 0.001703302012508331
     },
     {
       "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": 28203.653885699463,
-      "required_pages_per_second_upper_estimate": 0.0021711953488319762
+      "required_token_positions_per_second": 27971.24410976101,
+      "required_pages_per_second_upper_estimate": 0.0021533038009287978
     },
     {
       "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": 28203.653885699463,
-      "required_pages_per_second_upper_estimate": 0.35019498970286284
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27971.24410976101,
+      "required_pages_per_second_upper_estimate": 0.34730923811119124
     }
   ],
   "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": 0.06618685973448635,
-    "producer_p95_seconds": 0.2101595330004784,
-    "gpu_update_seconds": 0.9802985142297039,
-    "producer_batches_per_second": 15.10873916683125,
-    "gpu_batches_per_second": 1.0200974351019771,
-    "producer_headroom": 14.811074557128808,
+    "producer_batch_seconds": 0.08255714825001803,
+    "producer_p95_seconds": 0.2008701190006832,
+    "gpu_update_seconds": 0.9884436992329487,
+    "producer_batches_per_second": 12.112821496347914,
+    "gpu_batches_per_second": 1.011691410219944,
+    "producer_headroom": 11.972842087998513,
     "sustainable": true,
-    "buffer_seconds": 60.77850788224164,
+    "buffer_seconds": 60.295065653209875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517950.15515850723,
-    "upload_bytes_per_second": 6167111.21071734,
+    "generation_bytes_per_second": 522130.403813448,
+    "upload_bytes_per_second": 6022362.427711412,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.14004603449261,
+    "estimated_full_shards_per_hour": 112.03703008463458,
     "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": 1.078922188999968,
-      "bytes": 13377142,
-      "pages": 44,
-      "shards": 5
-    },
     {
       "seconds": 0.9435872209996887,
       "bytes": 6172084,
@@ -562,6 +556,12 @@
       "bytes": 11375746,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.7203741030007222,
+      "bytes": 1052913,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -136,3 +136,4 @@
 {"step": 53600, "loss": 0.6486603189259768, "lr": 0.0003, "grad_norm": 0.16541565818981485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7844871068, "tokens_per_second": 150020.08778311126, "data_wait_s": 1.011200038192328e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 53700, "loss": 0.6079864986240864, "lr": 0.0003, "grad_norm": 0.12471721482261457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7859579802, "tokens_per_second": 150323.1374425193, "data_wait_s": 9.87699968391098e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 53800, "loss": 0.6657801941037178, "lr": 0.0003, "grad_norm": 0.1372327896521534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7874295338, "tokens_per_second": 150565.55520741918, "data_wait_s": 9.418999979970977e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 53900, "loss": 0.64006482437253, "lr": 0.0003, "grad_norm": 0.12548678243748934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7888829524, "tokens_per_second": 148955.3998046398, "data_wait_s": 9.272000170312822e-06, "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 sha256:9b2bd2ccb1334e23248e1aa8f1d9d6a80db7cb89a990f471ed1e35fc1e49d75e
-size 294015
+oid sha256:ef24fd8ed0518bec55ef453f0f7cfb796a4170ba57d6bfebacd58818328de37e
+size 296133