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": 77244.66825524042,
-      "required_pages_per_second_upper_estimate": 0.00534396538222128
+      "required_token_positions_per_second": 77483.12669168787,
+      "required_pages_per_second_upper_estimate": 0.005360462490154517
     },
     {
       "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": 9655.583531905053,
-      "required_pages_per_second_upper_estimate": 0.004042591193296216
+      "required_token_positions_per_second": 9685.390836460983,
+      "required_pages_per_second_upper_estimate": 0.004055070889266484
     },
     {
       "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": 19311.167063810106,
-      "required_pages_per_second_upper_estimate": 0.004022006829669267
+      "required_token_positions_per_second": 19370.781672921967,
+      "required_pages_per_second_upper_estimate": 0.004034422980604303
     },
     {
       "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": 28966.75059571516,
-      "required_pages_per_second_upper_estimate": 0.0017365124685459448
+      "required_token_positions_per_second": 29056.17250938295,
+      "required_pages_per_second_upper_estimate": 0.0017418731757309726
     },
     {
       "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": 28966.75059571516,
-      "required_pages_per_second_upper_estimate": 0.002252037447546221
+      "required_token_positions_per_second": 29056.17250938295,
+      "required_pages_per_second_upper_estimate": 0.0022589896080084616
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28966.75059571516,
-      "required_pages_per_second_upper_estimate": 0.3715543745682477
+      "required_token_positions_per_second": 29056.17250938295,
+      "required_pages_per_second_upper_estimate": 0.37270138286300775
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3895362466718737,
-    "producer_p95_seconds": 0.19760117799978616,
-    "gpu_update_seconds": 0.954473644140457,
-    "producer_batches_per_second": 2.5671551968367945,
-    "gpu_batches_per_second": 1.0476978658751142,
-    "producer_headroom": 2.4502819757989274,
+    "producer_batch_seconds": 0.06548743603908491,
+    "producer_p95_seconds": 0.17932252299942775,
+    "gpu_update_seconds": 0.9515362008217629,
+    "producer_batches_per_second": 15.270104625918922,
+    "gpu_batches_per_second": 1.050932165414603,
+    "producer_headroom": 14.530057341897718,
     "sustainable": true,
-    "buffer_seconds": 60.13183958084879,
+    "buffer_seconds": 60.89831685259283,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386934.31705780013,
-    "upload_bytes_per_second": 2362602.022411542,
+    "generation_bytes_per_second": 377167.91065539676,
+    "upload_bytes_per_second": 2488202.0335134943,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.02709707069877,
+    "estimated_full_shards_per_hour": 80.9314536070483,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 2517815,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.2018508789997213,
+      "bytes": 11749429,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 651400, "loss": 0.5146275162696838, "lr": 0.0001, "grad_norm": 0.17616883334889957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95347576330, "tokens_per_second": 149689.76331282474, "data_wait_s": 1.3550999938161112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 651500, "loss": 0.5590850114822388, "lr": 0.0001, "grad_norm": 0.19672708800431543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95361996926, "tokens_per_second": 150293.79136367328, "data_wait_s": 1.3192999176681042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 651600, "loss": 0.5520280599594116, "lr": 0.0001, "grad_norm": 0.1837548457599139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95376651679, "tokens_per_second": 153053.669803133, "data_wait_s": 1.5702999917266425e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 651700, "loss": 0.5561076402664185, "lr": 0.0001, "grad_norm": 0.1972366777198106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95391310126, "tokens_per_second": 151506.7058284033, "data_wait_s": 9.306000720243901e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:717b19f49bb6523f1fd293b58652363f0fafc79498715261e875ec20bd8c13f4
-size 140736
+oid sha256:c5da3c345193cc3d958cbe0499bc8a25716973382643ec8530e5119643b0c452
+size 142854