Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74546.89354989669,
-      "required_pages_per_second_upper_estimate": 0.0064615357955306926
+      "required_token_positions_per_second": 75754.55234790196,
+      "required_pages_per_second_upper_estimate": 0.006566212599358549
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9318.361693737086,
-      "required_pages_per_second_upper_estimate": 0.003717233801554606
+      "required_token_positions_per_second": 9469.319043487745,
+      "required_pages_per_second_upper_estimate": 0.0037774529453836543
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18636.723387474172,
-      "required_pages_per_second_upper_estimate": 0.004311277960562955
+      "required_token_positions_per_second": 18938.63808697549,
+      "required_pages_per_second_upper_estimate": 0.004381120612775353
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27955.08508121126,
-      "required_pages_per_second_upper_estimate": 0.0017023180124495219
+      "required_token_positions_per_second": 28407.957130463234,
+      "required_pages_per_second_upper_estimate": 0.0017298955442129544
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27955.08508121126,
-      "required_pages_per_second_upper_estimate": 0.002152059834179984
+      "required_token_positions_per_second": 28407.957130463234,
+      "required_pages_per_second_upper_estimate": 0.0021869231781614694
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27955.08508121126,
-      "required_pages_per_second_upper_estimate": 0.347108597057393
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28407.957130463234,
+      "required_pages_per_second_upper_estimate": 0.35273175224385356
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.46083899110163884,
-    "producer_p95_seconds": 0.3825809999980265,
-    "gpu_update_seconds": 0.9890150546736253,
-    "producer_batches_per_second": 2.169955275723291,
-    "gpu_batches_per_second": 1.0111069546155693,
-    "producer_headroom": 2.146118435658792,
+    "producer_batch_seconds": 0.09591713410975444,
+    "producer_p95_seconds": 0.3745843260039692,
+    "gpu_update_seconds": 0.9732484413795354,
+    "producer_batches_per_second": 10.425665959282487,
+    "gpu_batches_per_second": 1.0274868753784445,
+    "producer_headroom": 10.14676314521536,
     "sustainable": true,
-    "buffer_seconds": 60.32991833509114,
+    "buffer_seconds": 60.34140336553119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367591.4107940038,
+    "generation_bytes_per_second": 366766.4687634059,
     "upload_bytes_per_second": 4234795.073533338,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.87655966630064,
+    "estimated_full_shards_per_hour": 78.69954631020197,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -343,3 +343,4 @@
 {"step": 34300, "loss": 0.5933354459702969, "lr": 0.0003, "grad_norm": 0.15678194327714132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5019240778, "tokens_per_second": 147608.11027876544, "data_wait_s": 1.2222000805195421e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34400, "loss": 0.6432075276970863, "lr": 0.0003, "grad_norm": 0.13552194736432954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5033696260, "tokens_per_second": 146893.72778329093, "data_wait_s": 1.0436997399665415e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34500, "loss": 0.6013185773044825, "lr": 0.0003, "grad_norm": 0.13190112491699676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5048308953, "tokens_per_second": 148703.76954503203, "data_wait_s": 1.3197000953368843e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 34600, "loss": 0.6039976011961699, "lr": 0.0003, "grad_norm": 0.1353956471650447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5063136574, "tokens_per_second": 151591.29069847715, "data_wait_s": 1.2245996913406998e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0df4890da2b2476517a6439d44b26a3f89d58b56445741cae59285418f340d48
-size 729799
+oid sha256:8d650a97e8ac6e79d7d85e98d39cbf3db2112d67fd6d9f6153c878fcb09223f4
+size 731917