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": 76763.54952824189,
-      "required_pages_per_second_upper_estimate": 0.005310680472338272
+      "required_token_positions_per_second": 76815.340863563,
+      "required_pages_per_second_upper_estimate": 0.005314263517088241
     },
     {
       "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": 9595.443691030236,
-      "required_pages_per_second_upper_estimate": 0.004017411897784616
+      "required_token_positions_per_second": 9601.917607945375,
+      "required_pages_per_second_upper_estimate": 0.004020122391606227
     },
     {
       "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": 19190.88738206047,
-      "required_pages_per_second_upper_estimate": 0.003996955743949333
+      "required_token_positions_per_second": 19203.83521589075,
+      "required_pages_per_second_upper_estimate": 0.00399965243627884
     },
     {
       "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": 28786.331073090707,
-      "required_pages_per_second_upper_estimate": 0.001725696593649269
+      "required_token_positions_per_second": 28805.752823836126,
+      "required_pages_per_second_upper_estimate": 0.0017268608979511607
     },
     {
       "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": 28786.331073090707,
-      "required_pages_per_second_upper_estimate": 0.002238010623243782
+      "required_token_positions_per_second": 28805.752823836126,
+      "required_pages_per_second_upper_estimate": 0.0022395205789369848
     },
     {
       "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": 28786.331073090707,
-      "required_pages_per_second_upper_estimate": 0.3692401466514117
+      "required_token_positions_per_second": 28805.752823836126,
+      "required_pages_per_second_upper_estimate": 0.3694892680165227
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11240627049238583,
-    "producer_p95_seconds": 0.35734798700286774,
-    "gpu_update_seconds": 0.9604558472491546,
-    "producer_batches_per_second": 8.896300852430986,
-    "gpu_batches_per_second": 1.0411722755024126,
-    "producer_headroom": 8.544504172604979,
+    "producer_batch_seconds": 0.11588609628188351,
+    "producer_p95_seconds": 0.36039401200105203,
+    "gpu_update_seconds": 0.9598082775021901,
+    "producer_batches_per_second": 8.629162877033853,
+    "gpu_batches_per_second": 1.0418747404454618,
+    "producer_headroom": 8.282341957291706,
     "sustainable": true,
-    "buffer_seconds": 60.508718376696734,
+    "buffer_seconds": 60.467921482637976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488548.9094728057,
-    "upload_bytes_per_second": 5424241.543300806,
+    "generation_bytes_per_second": 508704.4027182277,
+    "upload_bytes_per_second": 5334483.285516121,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.83122313631179,
+    "estimated_full_shards_per_hour": 109.15612279090999,
     "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": 1.4893900960014435,
-      "bytes": 11508997,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.178049504000228,
       "bytes": 7502276,
@@ -562,6 +556,12 @@
       "bytes": 11361644,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.925975751997612,
+      "bytes": 1221763,
+      "pages": 37,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -371,3 +371,4 @@
 {"step": 434200, "loss": 0.5645946264266968, "lr": 0.0003, "grad_norm": 0.16347681515306023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63554366941, "tokens_per_second": 152190.84287311233, "data_wait_s": 1.9464001525193453e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434300, "loss": 0.5650642514228821, "lr": 0.0003, "grad_norm": 0.13918712969755573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63568742841, "tokens_per_second": 148623.06905902762, "data_wait_s": 1.9120001525152475e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434400, "loss": 0.615990161895752, "lr": 0.0003, "grad_norm": 0.19471004069336245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63583297465, "tokens_per_second": 151005.618407158, "data_wait_s": 1.2661999789997935e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 434500, "loss": 0.5742749571800232, "lr": 0.0003, "grad_norm": 0.16708791504816872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63597931230, "tokens_per_second": 151603.94054082688, "data_wait_s": 1.2090000382158905e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8461ad1307a137e5ca9eebf79c14a3b9dd9b313400ae5d930c25eddbc1795468
-size 795660
+oid sha256:14e6a0ca2c67c68fbcc35b97dfd5d1cd714d1c714389e3e2c989dca43bfd6c85
+size 797778