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": 75845.09479637725,
-      "required_pages_per_second_upper_estimate": 0.0052471396428792885
+      "required_token_positions_per_second": 76407.9111220439,
+      "required_pages_per_second_upper_estimate": 0.0052860765822027025
     },
     {
       "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": 9480.636849547156,
-      "required_pages_per_second_upper_estimate": 0.003969344670695123
+      "required_token_positions_per_second": 9550.988890255487,
+      "required_pages_per_second_upper_estimate": 0.003998799601022032
     },
     {
       "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": 18961.273699094312,
-      "required_pages_per_second_upper_estimate": 0.003949133269107504
+      "required_token_positions_per_second": 19101.977780510973,
+      "required_pages_per_second_upper_estimate": 0.00397843821865044
     },
     {
       "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": 28441.91054864147,
-      "required_pages_per_second_upper_estimate": 0.0017050491091082276
+      "required_token_positions_per_second": 28652.966670766462,
+      "required_pages_per_second_upper_estimate": 0.001717701601365582
     },
     {
       "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": 28441.91054864147,
-      "required_pages_per_second_upper_estimate": 0.0022112334424136365
+      "required_token_positions_per_second": 28652.966670766462,
+      "required_pages_per_second_upper_estimate": 0.0022276421275710828
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28441.91054864147,
-      "required_pages_per_second_upper_estimate": 0.3648222899737237
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28652.966670766462,
+      "required_pages_per_second_upper_estimate": 0.3675294912939349
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.106011079359007,
-    "producer_p95_seconds": 0.3295924920021207,
-    "gpu_update_seconds": 0.9720865956847828,
-    "producer_batches_per_second": 9.432976308197896,
-    "gpu_batches_per_second": 1.028714935931766,
-    "producer_headroom": 9.169669826611303,
+    "producer_batch_seconds": 0.10879093923438177,
+    "producer_p95_seconds": 0.32612680799502414,
+    "gpu_update_seconds": 0.964926261133309,
+    "producer_batches_per_second": 9.191941967203505,
+    "gpu_batches_per_second": 1.0363486209044583,
+    "producer_headroom": 8.86954619496803,
     "sustainable": true,
-    "buffer_seconds": 60.26936893245654,
+    "buffer_seconds": 60.79035445139847,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 151,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 685510.9501720181,
-    "upload_bytes_per_second": 3888688.3999713254,
+    "generation_bytes_per_second": 666855.3792614836,
+    "upload_bytes_per_second": 3927796.2312513823,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.54734601435855,
+    "estimated_full_shards_per_hour": 71.54582039539042,
     "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": 1.5219117590022506,
-      "bytes": 2396663,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.5374675419989217,
       "bytes": 6254376,
@@ -562,6 +556,12 @@
       "bytes": 12961722,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.4560003849983332,
+      "bytes": 6473170,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 834700, "loss": 0.48385611176490784, "lr": 3e-05, "grad_norm": 0.28114723383564727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122176097630, "tokens_per_second": 150496.268789307, "data_wait_s": 9.189003321807832e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 834800, "loss": 0.5412554144859314, "lr": 3e-05, "grad_norm": 0.2698154589368708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122190688973, "tokens_per_second": 150956.17025571744, "data_wait_s": 7.982001989148557e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 834900, "loss": 0.5117660164833069, "lr": 3e-05, "grad_norm": 0.2857102331021943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122205454169, "tokens_per_second": 153142.0329584622, "data_wait_s": 2.237500302726403e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 835000, "loss": 0.4915982186794281, "lr": 3e-05, "grad_norm": 0.2849306084083531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122220139372, "tokens_per_second": 151828.51635509788, "data_wait_s": 1.1789998097810894e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29ec9e0f11ada6d1601ef63b8163b60e447183975433adcc7ecd11ac1df39860
-size 744045
+oid sha256:e0deda287a2909f878a4c08bdbe4a61db6f8aacf86e693cca1638658b9aa563b
+size 746163