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": 77060.14448798755,
-      "required_pages_per_second_upper_estimate": 0.005331199599848598
+      "required_token_positions_per_second": 77069.6996497935,
+      "required_pages_per_second_upper_estimate": 0.005331860648113364
     },
     {
       "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": 9632.518060998444,
-      "required_pages_per_second_upper_estimate": 0.0040329341622894225
+      "required_token_positions_per_second": 9633.712456224188,
+      "required_pages_per_second_upper_estimate": 0.004033434230628633
     },
     {
       "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": 19265.036121996887,
-      "required_pages_per_second_upper_estimate": 0.004012398971044319
+      "required_token_positions_per_second": 19267.424912448376,
+      "required_pages_per_second_upper_estimate": 0.0040128964930987235
     },
     {
       "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": 28897.55418299533,
-      "required_pages_per_second_upper_estimate": 0.0017323642492601957
+      "required_token_positions_per_second": 28901.137368672564,
+      "required_pages_per_second_upper_estimate": 0.0017325790557703382
     },
     {
       "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": 28897.55418299533,
-      "required_pages_per_second_upper_estimate": 0.002246657730820108
+      "required_token_positions_per_second": 28901.137368672564,
+      "required_pages_per_second_upper_estimate": 0.002246936307745745
     },
     {
       "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": 28897.55418299533,
-      "required_pages_per_second_upper_estimate": 0.3706667972831971
+      "required_token_positions_per_second": 28901.137368672564,
+      "required_pages_per_second_upper_estimate": 0.37071275854173963
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.13888229553117526,
-    "producer_p95_seconds": 0.363497867998376,
-    "gpu_update_seconds": 0.9567591715519431,
-    "producer_batches_per_second": 7.200341815890619,
-    "gpu_batches_per_second": 1.0451951021048658,
-    "producer_headroom": 6.888993070662322,
+    "producer_batch_seconds": 0.11592747762534827,
+    "producer_p95_seconds": 0.35787701699882746,
+    "gpu_update_seconds": 0.9566405517994976,
+    "producer_batches_per_second": 8.626082620651651,
+    "gpu_batches_per_second": 1.0453247022812704,
+    "producer_headroom": 8.252060438088252,
     "sustainable": true,
-    "buffer_seconds": 60.275827807772416,
+    "buffer_seconds": 60.26835476336835,
     "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": 360738.03359167324,
-    "upload_bytes_per_second": 4634033.565828152,
+    "generation_bytes_per_second": 353544.9060392489,
+    "upload_bytes_per_second": 4639888.393057113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.40598445713661,
+    "estimated_full_shards_per_hour": 75.86250673182583,
     "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": 1.4528134669999417,
-      "bytes": 6484124,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.400596306000807,
       "bytes": 7658131,
@@ -562,6 +556,12 @@
       "bytes": 12509791,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.2379947030058247,
+      "bytes": 6041226,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -404,3 +404,4 @@
 {"step": 644800, "loss": 0.5745224952697754, "lr": 0.0001, "grad_norm": 0.17889248196152127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94381940687, "tokens_per_second": 144931.13760888163, "data_wait_s": 7.866998203098774e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 644900, "loss": 0.48915520310401917, "lr": 0.0001, "grad_norm": 0.18857827802658592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94396571499, "tokens_per_second": 151698.3424844779, "data_wait_s": 1.4887998986523598e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645000, "loss": 0.5380465984344482, "lr": 0.0001, "grad_norm": 0.1812405295203612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94411289736, "tokens_per_second": 151737.97812397123, "data_wait_s": 1.7938000382855535e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 645100, "loss": 0.5672286152839661, "lr": 0.0001, "grad_norm": 0.19267931109074643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94425964227, "tokens_per_second": 143045.67401698208, "data_wait_s": 1.160599640570581e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98969fabfcd4eb238b950524b006c921b38704efa17adb723422dbe7bb8b05c8
-size 867120
+oid sha256:82d1f486f210ff7408761ae0fbf9bc6201e3e624809bddb32b42795172b1fa00
+size 869238