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": 76505.18146520831,
-      "required_pages_per_second_upper_estimate": 0.005292805970241108
+      "required_token_positions_per_second": 77508.40311093023,
+      "required_pages_per_second_upper_estimate": 0.005362211171487076
     },
     {
       "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": 9563.147683151039,
-      "required_pages_per_second_upper_estimate": 0.0040038902337029315
+      "required_token_positions_per_second": 9688.550388866279,
+      "required_pages_per_second_upper_estimate": 0.004056393727879624
     },
     {
       "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": 19126.295366302078,
-      "required_pages_per_second_upper_estimate": 0.003983502930472864
+      "required_token_positions_per_second": 19377.100777732558,
+      "required_pages_per_second_upper_estimate": 0.004035739083490349
     },
     {
       "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": 28689.443049453115,
-      "required_pages_per_second_upper_estimate": 0.0017198883045716415
+      "required_token_positions_per_second": 29065.65116659884,
+      "required_pages_per_second_upper_estimate": 0.0017424414067579938
     },
     {
       "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": 28689.443049453115,
-      "required_pages_per_second_upper_estimate": 0.0022304780055713365
+      "required_token_positions_per_second": 29065.65116659884,
+      "required_pages_per_second_upper_estimate": 0.00225972653191479
     },
     {
       "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": 28689.443049453115,
-      "required_pages_per_second_upper_estimate": 0.3679973711144433
+      "required_token_positions_per_second": 29065.65116659884,
+      "required_pages_per_second_upper_estimate": 0.372822964900384
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08670191942982797,
-    "producer_p95_seconds": 0.229220715998963,
-    "gpu_update_seconds": 0.9636994330054459,
-    "producer_batches_per_second": 11.533770031577538,
-    "gpu_batches_per_second": 1.0376679343696873,
-    "producer_headroom": 11.115087639846477,
+    "producer_batch_seconds": 0.06814466113269191,
+    "producer_p95_seconds": 0.23260312600177713,
+    "gpu_update_seconds": 0.9512258934619036,
+    "producer_batches_per_second": 14.674663919052893,
+    "gpu_batches_per_second": 1.051274998791914,
+    "producer_headroom": 13.958920297654247,
     "sustainable": true,
-    "buffer_seconds": 60.713064279343094,
+    "buffer_seconds": 60.87845718156183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356336.8037444865,
-    "upload_bytes_per_second": 5488469.345284643,
+    "generation_bytes_per_second": 355274.29141085007,
+    "upload_bytes_per_second": 5561972.705335598,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.46158298731753,
+    "estimated_full_shards_per_hour": 76.2335925745404,
     "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": 1.1028008130015223,
-      "bytes": 1043427,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 0.9759294559989939,
       "bytes": 7445734,
@@ -562,6 +556,12 @@
       "bytes": 2310644,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.012728962003166,
+      "bytes": 6259160,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 503400, "loss": 0.6006726622581482, "lr": 0.0003, "grad_norm": 0.16461212636053868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73681012188, "tokens_per_second": 152180.22700557596, "data_wait_s": 1.6964000678854063e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 503500, "loss": 0.540363073348999, "lr": 0.0003, "grad_norm": 0.18641936250339292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73695583004, "tokens_per_second": 151430.0645616692, "data_wait_s": 1.63710028573405e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 503600, "loss": 0.5512330532073975, "lr": 0.0003, "grad_norm": 0.17299385874191844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73710081336, "tokens_per_second": 151583.1382990878, "data_wait_s": 1.19440010166727e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 503700, "loss": 0.5739145278930664, "lr": 0.0003, "grad_norm": 0.14823518795035068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73724708395, "tokens_per_second": 116951.72401048958, "data_wait_s": 1.0966999980155379e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f62c08f60511e51caae33aef79fbaf9eb664dedf1cad2a503cdeadc91e3de395
-size 396462
+oid sha256:ca42c92d59a1b489321f246723da83b80215719e471a7cebccaab3f68a9ba690
+size 398580