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": 77067.38318634835,
-      "required_pages_per_second_upper_estimate": 0.0053317003897972925
+      "required_token_positions_per_second": 77088.40633828855,
+      "required_pages_per_second_upper_estimate": 0.005333154820228949
     },
     {
       "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": 9633.422898293544,
-      "required_pages_per_second_upper_estimate": 0.00403331299876973
+      "required_token_positions_per_second": 9636.050792286069,
+      "required_pages_per_second_upper_estimate": 0.004034413243107733
     },
     {
       "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": 19266.845796587088,
-      "required_pages_per_second_upper_estimate": 0.004012775878537129
+      "required_token_positions_per_second": 19272.101584572138,
+      "required_pages_per_second_upper_estimate": 0.004013870520569966
     },
     {
       "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": 28900.268694880633,
-      "required_pages_per_second_upper_estimate": 0.0017325269801029
+      "required_token_positions_per_second": 28908.152376858205,
+      "required_pages_per_second_upper_estimate": 0.00173299959479977
     },
     {
       "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": 28900.268694880633,
-      "required_pages_per_second_upper_estimate": 0.0022468687721793144
+      "required_token_positions_per_second": 28908.152376858205,
+      "required_pages_per_second_upper_estimate": 0.0022474816937764182
     },
     {
       "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": 28900.268694880633,
-      "required_pages_per_second_upper_estimate": 0.370701616127046
+      "required_token_positions_per_second": 28908.152376858205,
+      "required_pages_per_second_upper_estimate": 0.37080273953461607
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06606597767966349,
-    "producer_p95_seconds": 0.17650464400048804,
-    "gpu_update_seconds": 0.9566693061541515,
-    "producer_batches_per_second": 15.136383886555594,
-    "gpu_batches_per_second": 1.0452932832349766,
-    "producer_headroom": 14.480513870434018,
+    "producer_batch_seconds": 0.05555658316402656,
+    "producer_p95_seconds": 0.17854020100003254,
+    "gpu_update_seconds": 0.9564084082431019,
+    "producer_batches_per_second": 17.99966706101375,
+    "gpu_batches_per_second": 1.0455784279824294,
+    "producer_headroom": 17.215032922729954,
     "sustainable": true,
-    "buffer_seconds": 60.27016628771154,
+    "buffer_seconds": 60.25372971931542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334478.50115485786,
-    "upload_bytes_per_second": 4411451.492619816,
+    "generation_bytes_per_second": 333631.24687986355,
+    "upload_bytes_per_second": 4379228.935441446,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.77130008682539,
+    "estimated_full_shards_per_hour": 71.5894990424817,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 10655210,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 0.9483619619995807,
+      "bytes": 1257250,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -73,3 +73,4 @@
 {"step": 447500, "loss": 0.6167363524436951, "lr": 0.0003, "grad_norm": 0.18652145726225824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65500579390, "tokens_per_second": 151793.33238177557, "data_wait_s": 8.721999620320275e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 447600, "loss": 0.5296736359596252, "lr": 0.0003, "grad_norm": 0.1493937537363203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65515015591, "tokens_per_second": 150028.88444974553, "data_wait_s": 8.09299945103703e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 447700, "loss": 0.641356348991394, "lr": 0.0003, "grad_norm": 0.1681395916240247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65529471295, "tokens_per_second": 150984.10910820056, "data_wait_s": 1.4199999895936344e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 447800, "loss": 0.5502520203590393, "lr": 0.0003, "grad_norm": 0.1577973275191342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65544005141, "tokens_per_second": 151587.22604370053, "data_wait_s": 7.556999662483577e-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:b0bb029412a88124cf4253d27fbcd4b4c10d83bc7e6edc176e6c0096821296f5
-size 159798
+oid sha256:62a1a7a43682209cbb966bd11ec10e5eba70330d150dea8c3466547925190100
+size 161916