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": 77255.46968513484,
-      "required_pages_per_second_upper_estimate": 0.005344712650204146
+      "required_token_positions_per_second": 76092.81376448533,
+      "required_pages_per_second_upper_estimate": 0.005264277415880201
     },
     {
       "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": 9656.933710641855,
-      "required_pages_per_second_upper_estimate": 0.004043156484938377
+      "required_token_positions_per_second": 9511.601720560666,
+      "required_pages_per_second_upper_estimate": 0.003982309015568443
     },
     {
       "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": 19313.86742128371,
-      "required_pages_per_second_upper_estimate": 0.004022569242917803
+      "required_token_positions_per_second": 19023.20344112133,
+      "required_pages_per_second_upper_estimate": 0.003962031601174607
     },
     {
       "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": 28970.801131925567,
-      "required_pages_per_second_upper_estimate": 0.0017367552920069483
+      "required_token_positions_per_second": 28534.805161681998,
+      "required_pages_per_second_upper_estimate": 0.001710617999318146
     },
     {
       "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": 28970.801131925567,
-      "required_pages_per_second_upper_estimate": 0.0022523523589201544
+      "required_token_positions_per_second": 28534.805161681998,
+      "required_pages_per_second_upper_estimate": 0.002218455589977317
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28970.801131925567,
-      "required_pages_per_second_upper_estimate": 0.37160633049762787
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28534.805161681998,
+      "required_pages_per_second_upper_estimate": 0.36601384232734313
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06196427735150678,
-    "producer_p95_seconds": 0.21353033499872254,
-    "gpu_update_seconds": 0.9543401949465646,
-    "producer_batches_per_second": 16.138330708309038,
-    "gpu_batches_per_second": 1.0478443696442985,
-    "producer_headroom": 15.401457674279776,
+    "producer_batch_seconds": 0.07892034148436267,
+    "producer_p95_seconds": 0.2060119220004708,
+    "gpu_update_seconds": 0.9689219829377758,
+    "producer_batches_per_second": 12.671004473518918,
+    "gpu_batches_per_second": 1.0320748394705583,
+    "producer_headroom": 12.277214780295377,
     "sustainable": true,
-    "buffer_seconds": 60.12343228163357,
+    "buffer_seconds": 60.0731629421421,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345586.2764468814,
-    "upload_bytes_per_second": 5977036.19421587,
+    "generation_bytes_per_second": 346593.9961684603,
+    "upload_bytes_per_second": 5851802.473702633,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.15477008871872,
+    "estimated_full_shards_per_hour": 74.37100328245504,
     "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": 1.1030007570007,
-      "bytes": 11547889,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.05019463699864,
       "bytes": 6120696,
@@ -562,6 +556,12 @@
       "bytes": 7536660,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8494176380008867,
+      "bytes": 947075,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -138,3 +138,4 @@
 {"step": 498800, "loss": 0.5576951503753662, "lr": 0.0003, "grad_norm": 0.1574530003632909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73007594670, "tokens_per_second": 152748.16180447643, "data_wait_s": 1.1441999959060922e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498900, "loss": 0.6467173099517822, "lr": 0.0003, "grad_norm": 0.16527652480198224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73022166926, "tokens_per_second": 152253.54614836109, "data_wait_s": 1.338299989583902e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499000, "loss": 0.6090105772018433, "lr": 0.0003, "grad_norm": 0.16052634128681137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73036684139, "tokens_per_second": 151162.77354802587, "data_wait_s": 1.4625999028794467e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 499100, "loss": 0.5847358107566833, "lr": 0.0003, "grad_norm": 0.17837517196238098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73051376833, "tokens_per_second": 153421.47770888914, "data_wait_s": 1.0356998245697469e-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:c39aad1cfdc9612eb756fded259d055f018f881ed3621384bf37e99d0620ec76
-size 298251
+oid sha256:903fa7b78f5d951109384e17e704909cdb1dd44ab6a8528c60e198292c5f7088
+size 300369