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": 76965.16895985749,
-      "required_pages_per_second_upper_estimate": 0.005324628972438974
+      "required_token_positions_per_second": 76587.55062821397,
+      "required_pages_per_second_upper_estimate": 0.005298504460060629
     },
     {
       "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": 9620.646119982186,
-      "required_pages_per_second_upper_estimate": 0.004027963628500235
+      "required_token_positions_per_second": 9573.443828526746,
+      "required_pages_per_second_upper_estimate": 0.004008201014763776
     },
     {
       "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": 19241.292239964372,
-      "required_pages_per_second_upper_estimate": 0.004007453746585223
+      "required_token_positions_per_second": 19146.887657053492,
+      "required_pages_per_second_upper_estimate": 0.003987791761581158
     },
     {
       "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": 28861.93835994656,
-      "required_pages_per_second_upper_estimate": 0.0017302291350506347
+      "required_token_positions_per_second": 28720.33148558024,
+      "required_pages_per_second_upper_estimate": 0.0017217400191535514
     },
     {
       "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": 28861.93835994656,
-      "required_pages_per_second_upper_estimate": 0.0022438887572355155
+      "required_token_positions_per_second": 28720.33148558024,
+      "required_pages_per_second_upper_estimate": 0.002232879445616348
     },
     {
       "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": 28861.93835994656,
-      "required_pages_per_second_upper_estimate": 0.37020995574641885
+      "required_token_positions_per_second": 28720.33148558024,
+      "required_pages_per_second_upper_estimate": 0.36839357480766327
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08501760210934606,
-    "producer_p95_seconds": 0.1867994140002338,
-    "gpu_update_seconds": 0.9579398187049275,
-    "producer_batches_per_second": 11.762270108651643,
-    "gpu_batches_per_second": 1.043906914060567,
-    "producer_headroom": 11.267546895440143,
+    "producer_batch_seconds": 0.09815012935938938,
+    "producer_p95_seconds": 0.1959291100001792,
+    "gpu_update_seconds": 0.9626629836734778,
+    "producer_batches_per_second": 10.188473581510737,
+    "gpu_batches_per_second": 1.0387851376439612,
+    "producer_headroom": 9.80806637705553,
     "sustainable": true,
-    "buffer_seconds": 60.35020857841043,
+    "buffer_seconds": 60.647767971429104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532717.8964673753,
-    "upload_bytes_per_second": 3081967.294495208,
+    "generation_bytes_per_second": 533618.5061706988,
+    "upload_bytes_per_second": 3071346.0617332505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.30885954395242,
+    "estimated_full_shards_per_hour": 114.50210942116473,
     "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": [
@@ -340,6 +340,12 @@
       "bytes": 6284844,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9990272300001379,
+      "bytes": 2457257,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 692400, "loss": 0.5341150164604187, "lr": 8e-05, "grad_norm": 0.2055547347648032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101348716792, "tokens_per_second": 150329.32750909438, "data_wait_s": 1.503000021330081e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692500, "loss": 0.5633674263954163, "lr": 8e-05, "grad_norm": 0.22156870038852616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101363325064, "tokens_per_second": 152146.5813806953, "data_wait_s": 1.5280999832611997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692600, "loss": 0.5743834972381592, "lr": 8e-05, "grad_norm": 0.19790142158113494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101377989277, "tokens_per_second": 152639.34258261672, "data_wait_s": 1.735400019242661e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 692700, "loss": 0.5638164281845093, "lr": 8e-05, "grad_norm": 0.21678316106732431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101392716786, "tokens_per_second": 153700.706609766, "data_wait_s": 1.4446999557549134e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcf9cac65f7974cdc45ec62809d4a57f575d85b712ac5c6c16f0db65ab84626a
-size 161916
+oid sha256:f1845e81d9af048cabfb9ff0649bbee89760a29f504eb8175d525309697b8fa4
+size 164034