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": 74634.73872543132,
-      "required_pages_per_second_upper_estimate": 0.006469149992704472
+      "required_token_positions_per_second": 74992.09892109959,
+      "required_pages_per_second_upper_estimate": 0.006500125068743862
     },
     {
       "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": 9329.342340678915,
-      "required_pages_per_second_upper_estimate": 0.0037216141457950035
+      "required_token_positions_per_second": 9374.012365137449,
+      "required_pages_per_second_upper_estimate": 0.0037394336864278958
     },
     {
       "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": 18658.68468135783,
-      "required_pages_per_second_upper_estimate": 0.004316358319397367
+      "required_token_positions_per_second": 18748.024730274898,
+      "required_pages_per_second_upper_estimate": 0.004337025567383172
     },
     {
       "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": 27988.02702203675,
-      "required_pages_per_second_upper_estimate": 0.0017043240038127879
+      "required_token_positions_per_second": 28122.03709541235,
+      "required_pages_per_second_upper_estimate": 0.001712484514184845
     },
     {
       "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": 27988.02702203675,
-      "required_pages_per_second_upper_estimate": 0.002154595795973859
+      "required_token_positions_per_second": 28122.03709541235,
+      "required_pages_per_second_upper_estimate": 0.0021649122623859387
     },
     {
       "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": 27988.02702203675,
-      "required_pages_per_second_upper_estimate": 0.34751762571286177
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28122.03709541235,
+      "required_pages_per_second_upper_estimate": 0.3491815823213225
     }
   ],
   "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": 0.11618285833597497,
-    "producer_p95_seconds": 0.2647327539998514,
-    "gpu_update_seconds": 0.9878509827874247,
-    "producer_batches_per_second": 8.607121690088073,
-    "gpu_batches_per_second": 1.0122984310632503,
-    "producer_headroom": 8.502553620524463,
+    "producer_batch_seconds": 0.060862024211061794,
+    "producer_p95_seconds": 0.20391785200081358,
+    "gpu_update_seconds": 0.9831435719324302,
+    "producer_batches_per_second": 16.430606982970637,
+    "gpu_batches_per_second": 1.0171454389255046,
+    "producer_headroom": 16.153645638255686,
     "sustainable": true,
-    "buffer_seconds": 60.2589099500329,
+    "buffer_seconds": 60.95490145981067,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382562.46317608206,
-    "upload_bytes_per_second": 6029418.886385784,
+    "generation_bytes_per_second": 382436.3999771145,
+    "upload_bytes_per_second": 5966712.758392238,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.08899899923178,
+    "estimated_full_shards_per_hour": 82.06194877133443,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 17299657,
       "pages": 29,
       "shards": 5
+    },
+    {
+      "seconds": 0.8660049680001976,
+      "bytes": 1103360,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 126800, "loss": 0.6097234869375825, "lr": 0.0003, "grad_norm": 0.14419275058677364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18556838342, "tokens_per_second": 149293.96007045152, "data_wait_s": 1.1334001101204194e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126900, "loss": 0.6471667028963566, "lr": 0.0003, "grad_norm": 0.17106806992462145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18571312114, "tokens_per_second": 127553.6732559711, "data_wait_s": 1.4460998499998823e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127000, "loss": 0.5849825702607632, "lr": 0.0003, "grad_norm": 0.13022864424398561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18585914170, "tokens_per_second": 149260.8528617134, "data_wait_s": 1.6432000847999007e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 127100, "loss": 0.5815989589318633, "lr": 0.0003, "grad_norm": 0.13598667592178712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18600543329, "tokens_per_second": 149133.4866198227, "data_wait_s": 1.9907000023522414e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:758665a886afe1aa7210315b5f5faa9bc67c01b05fb3ff1734833f50dd61a1de
-size 255891
+oid sha256:d4f3f9ebeda8bd7122e1291fcc3d7ec45130d25340ded927d62e366542d67ad3
+size 258009