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": 77443.4463401675,
-      "required_pages_per_second_upper_estimate": 0.005357717311365237
+      "required_token_positions_per_second": 76347.51347884194,
+      "required_pages_per_second_upper_estimate": 0.005281898133104151
     },
     {
       "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": 9680.430792520938,
-      "required_pages_per_second_upper_estimate": 0.004052994222446283
+      "required_token_positions_per_second": 9543.439184855242,
+      "required_pages_per_second_upper_estimate": 0.003995638697026726
     },
     {
       "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": 19360.861585041876,
-      "required_pages_per_second_upper_estimate": 0.00403235688790919
+      "required_token_positions_per_second": 19086.878369710485,
+      "required_pages_per_second_upper_estimate": 0.003975293409578938
     },
     {
       "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": 29041.29237756281,
-      "required_pages_per_second_upper_estimate": 0.0017409811345490012
+      "required_token_positions_per_second": 28630.31755456573,
+      "required_pages_per_second_upper_estimate": 0.001716343821432546
     },
     {
       "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": 29041.29237756281,
-      "required_pages_per_second_upper_estimate": 0.0022578327432102286
+      "required_token_positions_per_second": 28630.31755456573,
+      "required_pages_per_second_upper_estimate": 0.0022258812584094096
     },
     {
       "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": 29041.29237756281,
-      "required_pages_per_second_upper_estimate": 0.3725105165090598
+      "required_token_positions_per_second": 28630.31755456573,
+      "required_pages_per_second_upper_estimate": 0.3672389727500382
     }
   ],
   "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": 0.08359366537501955,
-    "producer_p95_seconds": 0.19113960699905874,
-    "gpu_update_seconds": 0.952023747447298,
-    "producer_batches_per_second": 11.962628932632398,
-    "gpu_batches_per_second": 1.0503939662023587,
-    "producer_headroom": 11.388706825766166,
+    "producer_batch_seconds": 0.629585273656204,
+    "producer_p95_seconds": 0.1781118359995162,
+    "gpu_update_seconds": 0.9656896032433606,
+    "producer_batches_per_second": 1.5883471895596901,
+    "gpu_batches_per_second": 1.035529425440022,
+    "producer_headroom": 1.5338503672986041,
     "sustainable": true,
-    "buffer_seconds": 60.92951983662707,
+    "buffer_seconds": 60.83844500433172,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347548.10059185437,
-    "upload_bytes_per_second": 4112177.086591685,
+    "generation_bytes_per_second": 291738.8336736196,
+    "upload_bytes_per_second": 4135218.9557089163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.57573188130115,
+    "estimated_full_shards_per_hour": 62.60036237389031,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 2361558,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0782337069995265,
+      "bytes": 6047629,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 934300, "loss": 0.5023261308670044, "lr": 3e-05, "grad_norm": 0.3236470582941519, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136755278901, "tokens_per_second": 153840.84962801982, "data_wait_s": 1.406699993822258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 934400, "loss": 0.5163746476173401, "lr": 3e-05, "grad_norm": 0.3394334857820853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136769953443, "tokens_per_second": 152630.65066769198, "data_wait_s": 1.952499951585196e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 934500, "loss": 0.48397529125213623, "lr": 3e-05, "grad_norm": 0.3178558022634129, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136784491604, "tokens_per_second": 148289.76532684793, "data_wait_s": 1.2567999874590896e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 934600, "loss": 0.4541608393192291, "lr": 3e-05, "grad_norm": 0.39213525667573657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136798980793, "tokens_per_second": 151881.48298382616, "data_wait_s": 6.8280005507403985e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6744d96799b4b7ca337cc5c949c9372413773fd13fd1d9a9f4349af6ad9bf0dd
-size 202158
+oid sha256:5ac6ead9984616ab93417ced556d1219eceb64a3271487e8d5ef264dcb83c6e7
+size 204276