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": 77686.07267619061,
-      "required_pages_per_second_upper_estimate": 0.005374502764261977
+      "required_token_positions_per_second": 77282.93503130684,
+      "required_pages_per_second_upper_estimate": 0.005346612766580708
     },
     {
       "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": 9710.759084523826,
-      "required_pages_per_second_upper_estimate": 0.004065692044981137
+      "required_token_positions_per_second": 9660.366878913355,
+      "required_pages_per_second_upper_estimate": 0.004044593880800948
     },
     {
       "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": 19421.518169047653,
-      "required_pages_per_second_upper_estimate": 0.004044990054735903
+      "required_token_positions_per_second": 19320.73375782671,
+      "required_pages_per_second_upper_estimate": 0.004023999319742238
     },
     {
       "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": 29132.27725357148,
-      "required_pages_per_second_upper_estimate": 0.0017464355389398593
+      "required_token_positions_per_second": 28981.100636740066,
+      "required_pages_per_second_upper_estimate": 0.0017373727315941443
     },
     {
       "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": 29132.27725357148,
-      "required_pages_per_second_upper_estimate": 0.0022649064171197278
+      "required_token_positions_per_second": 28981.100636740066,
+      "required_pages_per_second_upper_estimate": 0.0022531531001167477
     },
     {
       "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": 29132.27725357148,
-      "required_pages_per_second_upper_estimate": 0.37367757280655045
+      "required_token_positions_per_second": 28981.100636740066,
+      "required_pages_per_second_upper_estimate": 0.3717384414866416
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05681737146881005,
-    "producer_p95_seconds": 0.18071310099912807,
-    "gpu_update_seconds": 0.9490504212680623,
-    "producer_batches_per_second": 17.600251017401447,
-    "gpu_batches_per_second": 1.053684796497811,
-    "producer_headroom": 16.703525642488483,
+    "producer_batch_seconds": 0.09876583939851002,
+    "producer_p95_seconds": 0.20172957000067981,
+    "gpu_update_seconds": 0.9540010349003081,
+    "producer_batches_per_second": 10.124958245584311,
+    "gpu_batches_per_second": 1.0482168922432027,
+    "producer_headroom": 9.659220644609842,
     "sustainable": true,
-    "buffer_seconds": 60.739226961155985,
+    "buffer_seconds": 60.10206519871941,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354606.1178936668,
-    "upload_bytes_per_second": 3409288.2780103944,
+    "generation_bytes_per_second": 354393.6735847763,
+    "upload_bytes_per_second": 3405505.479165941,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.09021809203627,
+    "estimated_full_shards_per_hour": 76.04463248879878,
     "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": [
@@ -496,6 +496,18 @@
       "bytes": 6175312,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5478082320005342,
+      "bytes": 6142048,
+      "pages": 37,
+      "shards": 2
+    },
+    {
+      "seconds": 0.98161319600149,
+      "bytes": 2080200,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"step": 654800, "loss": 0.5525216460227966, "lr": 0.0001, "grad_norm": 0.1807556533966403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95845512701, "tokens_per_second": 152991.05662446775, "data_wait_s": 8.007998985704035e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 654900, "loss": 0.5827817320823669, "lr": 0.0001, "grad_norm": 0.1737878096522394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95860166615, "tokens_per_second": 153401.84862541239, "data_wait_s": 8.524999429937452e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 655000, "loss": 0.4905214011669159, "lr": 0.0001, "grad_norm": 0.17499976726772343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95874859680, "tokens_per_second": 153183.04710755683, "data_wait_s": 8.604998583905399e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 655100, "loss": 0.5905359387397766, "lr": 0.0001, "grad_norm": 0.17772512174507585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95889298486, "tokens_per_second": 140685.0785457286, "data_wait_s": 8.76400008564815e-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 sha256:16bb44394d89683ec21017d6bf3762794d670be9316b6f6fe182f5798e8215cd
-size 213531
+oid sha256:a01fe6421483c6f3badcbf5bebb306e384c9d4079a247f0811e364069a4407c5
+size 215649