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": 76043.69898119541,
-      "required_pages_per_second_upper_estimate": 0.005260879541210207
+      "required_token_positions_per_second": 76262.94248325225,
+      "required_pages_per_second_upper_estimate": 0.005276047315397511
     },
     {
       "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": 9505.462372649426,
-      "required_pages_per_second_upper_estimate": 0.00397973859880217
+      "required_token_positions_per_second": 9532.867810406531,
+      "required_pages_per_second_upper_estimate": 0.003991212683300452
     },
     {
       "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": 19010.924745298853,
-      "required_pages_per_second_upper_estimate": 0.003959474272645769
+      "required_token_positions_per_second": 19065.735620813062,
+      "required_pages_per_second_upper_estimate": 0.003970889932555389
     },
     {
       "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": 28516.38711794828,
-      "required_pages_per_second_upper_estimate": 0.001709513865719034
+      "required_token_positions_per_second": 28598.60343121959,
+      "required_pages_per_second_upper_estimate": 0.0017144426081626076
     },
     {
       "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": 28516.38711794828,
-      "required_pages_per_second_upper_estimate": 0.002217023668089433
+      "required_token_positions_per_second": 28598.60343121959,
+      "required_pages_per_second_upper_estimate": 0.0022234156248149336
     },
     {
       "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": 28516.38711794828,
-      "required_pages_per_second_upper_estimate": 0.36577759543808164
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28598.60343121959,
+      "required_pages_per_second_upper_estimate": 0.36683217802772655
     }
   ],
   "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": 0.0851213738123704,
-    "producer_p95_seconds": 0.2423511220004002,
-    "gpu_update_seconds": 0.9695477861779441,
-    "producer_batches_per_second": 11.747930692522182,
-    "gpu_batches_per_second": 1.0314086775878284,
-    "producer_headroom": 11.390180195106803,
+    "producer_batch_seconds": 0.5778800869842087,
+    "producer_p95_seconds": 0.2511357590010448,
+    "gpu_update_seconds": 0.9667604946687058,
+    "producer_batches_per_second": 1.730462811443659,
+    "gpu_batches_per_second": 1.0343823579000142,
+    "producer_headroom": 1.6729430835970713,
     "sustainable": true,
-    "buffer_seconds": 60.111962743032535,
+    "buffer_seconds": 60.90591116412847,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429193.8358567233,
-    "upload_bytes_per_second": 3859740.958862179,
+    "generation_bytes_per_second": 370918.6275528217,
+    "upload_bytes_per_second": 3854907.7736047916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.0950060536983,
+    "estimated_full_shards_per_hour": 79.59050292910088,
     "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": 1.5977993900014553,
-      "bytes": 5977865,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.8666853409995383,
       "bytes": 6845301,
@@ -562,6 +556,12 @@
       "bytes": 5860537,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.8819817990006413,
+      "bytes": 6541422,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -157,3 +157,4 @@
 {"step": 815700, "loss": 0.5718827247619629, "lr": 3e-05, "grad_norm": 0.2807549818283929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119395443058, "tokens_per_second": 151021.98915335568, "data_wait_s": 1.8255999748362228e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815800, "loss": 0.5310738682746887, "lr": 3e-05, "grad_norm": 0.27034295817646703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119409939459, "tokens_per_second": 149721.6236183282, "data_wait_s": 1.57389986270573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815900, "loss": 0.5196565985679626, "lr": 3e-05, "grad_norm": 0.2750242557713058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119424498157, "tokens_per_second": 128133.5540056803, "data_wait_s": 1.5623998478986323e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 816000, "loss": 0.5105184316635132, "lr": 3e-05, "grad_norm": 0.2745735589355725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119439089039, "tokens_per_second": 151066.15797413554, "data_wait_s": 1.1371997970854864e-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:bd23f7164a8a424912cb1e0a57973be4b7a8bf3b52f0f235bebd0b4d96fc5e51
-size 339276
+oid sha256:f7731cd20e1bf4da2d392ed46321abfe0056b1e97a66c747f1ea4ee7f2d8a475
+size 341394