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": 75513.34139560419,
-      "required_pages_per_second_upper_estimate": 0.006545305045356948
+      "required_token_positions_per_second": 75372.77811391844,
+      "required_pages_per_second_upper_estimate": 0.006533121376354807
     },
     {
       "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": 9439.167674450524,
-      "required_pages_per_second_upper_estimate": 0.003765425113471567
+      "required_token_positions_per_second": 9421.597264239805,
+      "required_pages_per_second_upper_estimate": 0.0037584160141374678
     },
     {
       "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": 18878.335348901048,
-      "required_pages_per_second_upper_estimate": 0.004367170635613773
+      "required_token_positions_per_second": 18843.19452847961,
+      "required_pages_per_second_upper_estimate": 0.0043590414252665894
     },
     {
       "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": 28317.50302335157,
-      "required_pages_per_second_upper_estimate": 0.0017243873636658771
+      "required_token_positions_per_second": 28264.791792719414,
+      "required_pages_per_second_upper_estimate": 0.0017211775262748325
     },
     {
       "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": 28317.50302335157,
-      "required_pages_per_second_upper_estimate": 0.0021799597706030193
+      "required_token_positions_per_second": 28264.791792719414,
+      "required_pages_per_second_upper_estimate": 0.0021759019141549306
     },
     {
       "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": 28317.50302335157,
-      "required_pages_per_second_upper_estimate": 0.35160861496394913
+      "required_token_positions_per_second": 28264.791792719414,
+      "required_pages_per_second_upper_estimate": 0.35095411789263836
     }
   ],
   "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": 0.2436919745156203,
-    "producer_p95_seconds": 0.1566531689995827,
-    "gpu_update_seconds": 0.9763572719388613,
-    "producer_batches_per_second": 4.103540963906062,
-    "gpu_batches_per_second": 1.024215242453399,
-    "producer_headroom": 4.006522060808688,
+    "producer_batch_seconds": 0.1419702244843677,
+    "producer_p95_seconds": 0.15981886500003384,
+    "gpu_update_seconds": 0.9781780882292475,
+    "producer_batches_per_second": 7.043730497940501,
+    "gpu_batches_per_second": 1.0223087309287981,
+    "producer_headroom": 6.890022832477485,
     "sustainable": true,
-    "buffer_seconds": 60.5341508602094,
+    "buffer_seconds": 60.647041470213345,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 161,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539578.2401181856,
-    "upload_bytes_per_second": 2888640.3101417464,
+    "generation_bytes_per_second": 567472.0381941635,
+    "upload_bytes_per_second": 2996380.5200971365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.78092959078957,
+    "estimated_full_shards_per_hour": 60.883144661754024,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 9860015,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8969114139999874,
+      "bytes": 7503017,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -19,3 +19,4 @@
 {"step": 1900, "loss": 0.9189868532121181, "lr": 0.0003, "grad_norm": 0.23279412071070146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 278648795, "tokens_per_second": 148790.8751210036, "data_wait_s": 1.0176000159844989e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 2000, "loss": 0.9953780174255371, "lr": 0.0003, "grad_norm": 0.2292280217590217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 293112776, "tokens_per_second": 146925.25477945834, "data_wait_s": 1.2533999779407168e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 2100, "loss": 0.8481050059199333, "lr": 0.0003, "grad_norm": 0.23960433516464869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 307689242, "tokens_per_second": 147659.70686422856, "data_wait_s": 1.485599977968377e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 2200, "loss": 0.8594202846288681, "lr": 0.0003, "grad_norm": 0.21901127440206275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 322336648, "tokens_per_second": 149103.51465016266, "data_wait_s": 1.2503000107244588e-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 sha256:5a79fbfba275a5edbe8e04dfb210a99b50773bab086f703735bbe59b95e2bcb3
-size 43869
+oid sha256:fac8ca34f5022a1510905a7670d93c1fc11a4c63fda2b99c993758fcbf1a1687
+size 45952