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": 74992.09892109959,
-      "required_pages_per_second_upper_estimate": 0.006500125068743862
+      "required_token_positions_per_second": 74987.4089146153,
+      "required_pages_per_second_upper_estimate": 0.006499718550868516
     },
     {
       "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": 9374.012365137449,
-      "required_pages_per_second_upper_estimate": 0.0037394336864278958
+      "required_token_positions_per_second": 9373.426114326912,
+      "required_pages_per_second_upper_estimate": 0.003739199822214342
     },
     {
       "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": 18748.024730274898,
-      "required_pages_per_second_upper_estimate": 0.004337025567383172
+      "required_token_positions_per_second": 18746.852228653825,
+      "required_pages_per_second_upper_estimate": 0.004336754329768461
     },
     {
       "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": 28122.03709541235,
-      "required_pages_per_second_upper_estimate": 0.001712484514184845
+      "required_token_positions_per_second": 28120.27834298074,
+      "required_pages_per_second_upper_estimate": 0.001712377415389221
     },
     {
       "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": 28122.03709541235,
-      "required_pages_per_second_upper_estimate": 0.0021649122623859387
+      "required_token_positions_per_second": 28120.27834298074,
+      "required_pages_per_second_upper_estimate": 0.0021647768687552147
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28122.03709541235,
-      "required_pages_per_second_upper_estimate": 0.3491815823213225
+      "required_token_positions_per_second": 28120.27834298074,
+      "required_pages_per_second_upper_estimate": 0.34915974450228765
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.0804506813436916,
+    "producer_p95_seconds": 0.2353090409997094,
+    "gpu_update_seconds": 0.9832050615850811,
+    "producer_batches_per_second": 12.42997552410926,
+    "gpu_batches_per_second": 1.017081826641375,
+    "producer_headroom": 12.221214850682896,
     "sustainable": true,
-    "buffer_seconds": 60.95490145981067,
+    "buffer_seconds": 60.95871381827503,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382436.3999771145,
-    "upload_bytes_per_second": 5966712.758392238,
+    "generation_bytes_per_second": 382732.9072604324,
+    "upload_bytes_per_second": 5970183.370832096,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.06194877133443,
+    "estimated_full_shards_per_hour": 82.12557233199814,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 1103360,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 0.9722155079998629,
+      "bytes": 6032233,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -119,3 +119,4 @@
 {"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}
+{"step": 127200, "loss": 0.6274984590709209, "lr": 0.0003, "grad_norm": 0.13884664794655535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18615248565, "tokens_per_second": 149852.5159162955, "data_wait_s": 2.2731999706593342e-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:d4f3f9ebeda8bd7122e1291fcc3d7ec45130d25340ded927d62e366542d67ad3
-size 258009
+oid sha256:8a52a30d12727b211683bcbbd544e7f967ba004856ab14c2643c7e1ad017e9ba
+size 260127