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": 77299.45470547979,
-      "required_pages_per_second_upper_estimate": 0.0053477556359709175
+      "required_token_positions_per_second": 76818.99777550212,
+      "required_pages_per_second_upper_estimate": 0.005314516510741394
     },
     {
       "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": 9662.431838184973,
-      "required_pages_per_second_upper_estimate": 0.004045458436126722
+      "required_token_positions_per_second": 9602.374721937766,
+      "required_pages_per_second_upper_estimate": 0.004020313775689215
     },
     {
       "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": 19324.863676369947,
-      "required_pages_per_second_upper_estimate": 0.004024859472861519
+      "required_token_positions_per_second": 19204.74944387553,
+      "required_pages_per_second_upper_estimate": 0.003999842845858257
     },
     {
       "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": 28987.29551455492,
-      "required_pages_per_second_upper_estimate": 0.0017377441050601139
+      "required_token_positions_per_second": 28807.124165813297,
+      "required_pages_per_second_upper_estimate": 0.001726943107808774
     },
     {
       "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": 28987.29551455492,
-      "required_pages_per_second_upper_estimate": 0.002253634724618465
+      "required_token_positions_per_second": 28807.124165813297,
+      "required_pages_per_second_upper_estimate": 0.002239627194743814
     },
     {
       "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": 28987.29551455492,
-      "required_pages_per_second_upper_estimate": 0.371817902727709
+      "required_token_positions_per_second": 28807.124165813297,
+      "required_pages_per_second_upper_estimate": 0.36950685811897355
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0835076106716599,
-    "producer_p95_seconds": 0.23780048499975237,
-    "gpu_update_seconds": 0.9537971552440123,
-    "producer_batches_per_second": 11.97495643758577,
-    "gpu_batches_per_second": 1.048440954664168,
-    "producer_headroom": 11.42167938434028,
+    "producer_batch_seconds": 0.10149270860136994,
+    "producer_p95_seconds": 0.25899350799954846,
+    "gpu_update_seconds": 0.9597625865344489,
+    "producer_batches_per_second": 9.852924547788668,
+    "gpu_batches_per_second": 1.041924340488039,
+    "producer_headroom": 9.456468348914418,
     "sustainable": true,
-    "buffer_seconds": 60.08922078037278,
+    "buffer_seconds": 60.465042951670284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280989.70904241805,
-    "upload_bytes_per_second": 5016201.149272126,
+    "generation_bytes_per_second": 314474.70619697584,
+    "upload_bytes_per_second": 4903172.445268796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.2938504548493,
+    "estimated_full_shards_per_hour": 67.47895135337788,
     "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": [
-    {
-      "seconds": 1.1364503580007295,
-      "bytes": 11698392,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0955314789989643,
       "bytes": 11552086,
@@ -562,6 +556,12 @@
       "bytes": 6145347,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.223426783002651,
+      "bytes": 2322033,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 462500, "loss": 0.6043637990951538, "lr": 0.0003, "grad_norm": 0.19626290176222383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67696317791, "tokens_per_second": 152813.25386393775, "data_wait_s": 8.059996616793796e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 462600, "loss": 0.5018669366836548, "lr": 0.0003, "grad_norm": 0.1580707417784199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67710761565, "tokens_per_second": 149731.8577879642, "data_wait_s": 7.398000889224932e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 462700, "loss": 0.5562172532081604, "lr": 0.0003, "grad_norm": 0.15023338817400747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67725234326, "tokens_per_second": 151000.43977930836, "data_wait_s": 1.2661999789997935e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 462800, "loss": 0.5868706107139587, "lr": 0.0003, "grad_norm": 0.16302416748524298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67739832563, "tokens_per_second": 151067.1121372493, "data_wait_s": 7.637001544935629e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99c855f18055099ac1560b7d3447ddd838198c5db8db30a77bac92331d113970
-size 479847
+oid sha256:7e9e610d9d7c50e91f8ce795c154e5118da2e054dd1ca488d90bf564d4ea3e7d
+size 481965