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": 74748.73762680705,
-      "required_pages_per_second_upper_estimate": 0.005171291109989308
+      "required_token_positions_per_second": 75790.72568758717,
+      "required_pages_per_second_upper_estimate": 0.005243378261779477
     },
     {
       "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": 9343.592203350881,
-      "required_pages_per_second_upper_estimate": 0.0039119669391503836
+      "required_token_positions_per_second": 9473.840710948396,
+      "required_pages_per_second_upper_estimate": 0.003966499269383334
     },
     {
       "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": 18687.184406701763,
-      "required_pages_per_second_upper_estimate": 0.0038920476977228535
+      "required_token_positions_per_second": 18947.681421896792,
+      "required_pages_per_second_upper_estimate": 0.003946302356219716
     },
     {
       "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": 28030.776610052642,
-      "required_pages_per_second_upper_estimate": 0.0016804022572550028
+      "required_token_positions_per_second": 28421.52213284519,
+      "required_pages_per_second_upper_estimate": 0.0017038268547125489
     },
     {
       "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": 28030.776610052642,
-      "required_pages_per_second_upper_estimate": 0.0021792695870754344
+      "required_token_positions_per_second": 28421.52213284519,
+      "required_pages_per_second_upper_estimate": 0.002209648332764639
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28030.776610052642,
-      "required_pages_per_second_upper_estimate": 0.3595487052507362
+      "required_token_positions_per_second": 28421.52213284519,
+      "required_pages_per_second_upper_estimate": 0.3645607692672643
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2629237157809996,
-    "producer_p95_seconds": 0.2750178429996595,
-    "gpu_update_seconds": 0.9863444165184004,
-    "producer_batches_per_second": 3.8033845559711423,
-    "gpu_batches_per_second": 1.0138446401205383,
-    "producer_headroom": 3.751447120654452,
+    "producer_batch_seconds": 0.09397501781228357,
+    "producer_p95_seconds": 0.282618855999317,
+    "gpu_update_seconds": 0.9727839301065697,
+    "producer_batches_per_second": 10.641125942615028,
+    "gpu_batches_per_second": 1.0279775076983937,
+    "producer_headroom": 10.351516315216022,
     "sustainable": true,
-    "buffer_seconds": 60.16700940762242,
+    "buffer_seconds": 60.31260366660732,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336132.43174712,
-    "upload_bytes_per_second": 5152756.524796958,
+    "generation_bytes_per_second": 338497.3711533929,
+    "upload_bytes_per_second": 5172575.955089351,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.12619509039116,
+    "estimated_full_shards_per_hour": 72.63365603400554,
     "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": 1.3693337599997903,
-      "bytes": 10798536,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.092897812000956,
       "bytes": 6537422,
@@ -562,6 +556,12 @@
       "bytes": 7369637,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1391658569991705,
+      "bytes": 11247930,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 740000, "loss": 0.5588091611862183, "lr": 5e-05, "grad_norm": 0.2407953392202154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108315982560, "tokens_per_second": 150483.63865526207, "data_wait_s": 1.4635999832535163e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 740100, "loss": 0.5209000110626221, "lr": 5e-05, "grad_norm": 0.2159172759914041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108330641559, "tokens_per_second": 139892.70958568153, "data_wait_s": 1.8845999875338748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 740200, "loss": 0.5307397842407227, "lr": 5e-05, "grad_norm": 0.23519545309184453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108345251095, "tokens_per_second": 150095.86339141673, "data_wait_s": 1.8558999727247283e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 740300, "loss": 0.6049389839172363, "lr": 5e-05, "grad_norm": 0.2738369921934624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108359961977, "tokens_per_second": 151296.16616417596, "data_wait_s": 1.8677998014027253e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dae27c1de48cb92cd846fc7183ad21b1a8567e483162d791588f400ea5370927
-size 431133
+oid sha256:228b58376eebe9e8d0cb437c21e2b78d91c2fd11e6ca8bf12347cf043991f699
+size 433251