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": 76458.74483657874,
-      "required_pages_per_second_upper_estimate": 0.005289593376524691
+      "required_token_positions_per_second": 76087.46368546286,
+      "required_pages_per_second_upper_estimate": 0.0052639072849995335
     },
     {
       "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": 9557.343104572343,
-      "required_pages_per_second_upper_estimate": 0.004001459977865416
+      "required_token_positions_per_second": 9510.932960682858,
+      "required_pages_per_second_upper_estimate": 0.003982029019772899
     },
     {
       "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": 19114.686209144686,
-      "required_pages_per_second_upper_estimate": 0.0039810850491910415
+      "required_token_positions_per_second": 19021.865921365716,
+      "required_pages_per_second_upper_estimate": 0.00396175303108228
     },
     {
       "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": 28672.02931371703,
-      "required_pages_per_second_upper_estimate": 0.001718844377703498
+      "required_token_positions_per_second": 28532.798882048573,
+      "required_pages_per_second_upper_estimate": 0.0017104977259175338
     },
     {
       "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": 28672.02931371703,
-      "required_pages_per_second_upper_estimate": 0.0022291241642127847
+      "required_token_positions_per_second": 28532.798882048573,
+      "required_pages_per_second_upper_estimate": 0.0022182996105605097
     },
     {
       "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": 28672.02931371703,
-      "required_pages_per_second_upper_estimate": 0.3677740064098335
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28532.798882048573,
+      "required_pages_per_second_upper_estimate": 0.3659881079263808
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08217706103121714,
-    "producer_p95_seconds": 0.19257245599874295,
-    "gpu_update_seconds": 0.9642847284190268,
-    "producer_batches_per_second": 12.168845994871043,
-    "gpu_batches_per_second": 1.0370380972843254,
-    "producer_headroom": 11.734232355337186,
+    "producer_batch_seconds": 0.05395135758601555,
+    "producer_p95_seconds": 0.17744025899992266,
+    "gpu_update_seconds": 0.9689901125470994,
+    "producer_batches_per_second": 18.535214770187817,
+    "gpu_batches_per_second": 1.0320022743796504,
+    "producer_headroom": 17.960439846248953,
     "sustainable": true,
-    "buffer_seconds": 60.74993789039869,
+    "buffer_seconds": 60.07738697792016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285119.93521664024,
-    "upload_bytes_per_second": 4061027.5664061466,
+    "generation_bytes_per_second": 305430.6522120692,
+    "upload_bytes_per_second": 4106818.974489624,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.180100845092824,
+    "estimated_full_shards_per_hour": 65.53830790301853,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 7212989,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.7552309799993964,
+      "bytes": 9939009,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -85,3 +85,4 @@
 {"step": 808500, "loss": 0.5446516871452332, "lr": 3e-05, "grad_norm": 0.26892138771215574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118341033127, "tokens_per_second": 152401.82121693698, "data_wait_s": 1.0585001291474327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 808600, "loss": 0.5144476294517517, "lr": 3e-05, "grad_norm": 0.2696697976138345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118355747682, "tokens_per_second": 145440.98551701644, "data_wait_s": 9.385001249029301e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 808700, "loss": 0.4755537509918213, "lr": 3e-05, "grad_norm": 0.27892815376068164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118370460749, "tokens_per_second": 152427.4103789687, "data_wait_s": 1.0581999958958477e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 808800, "loss": 0.5579580664634705, "lr": 3e-05, "grad_norm": 0.3085535286837493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118384829091, "tokens_per_second": 149086.49574901848, "data_wait_s": 1.5311999959521927e-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 sha256:2137606d1f42556ec2636dd22d55b29676bdde5f18ebe119e860b8d93ea6a11b
-size 185214
+oid sha256:be82c939ee77f0b618aa645c06474bdd745c1dc5b01673db4dec93664e648ef8
+size 187332