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": 76482.32704317285,
-      "required_pages_per_second_upper_estimate": 0.005291224848295129
+      "required_token_positions_per_second": 76460.77215939954,
+      "required_pages_per_second_upper_estimate": 0.005289733631421471
     },
     {
       "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": 9560.290880396606,
-      "required_pages_per_second_upper_estimate": 0.0040026941500464755
+      "required_token_positions_per_second": 9557.596519924942,
+      "required_pages_per_second_upper_estimate": 0.004001566077581635
     },
     {
       "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": 19120.58176079321,
-      "required_pages_per_second_upper_estimate": 0.003982312937123277
+      "required_token_positions_per_second": 19115.193039849884,
+      "required_pages_per_second_upper_estimate": 0.00398119060866091
     },
     {
       "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": 28680.872641189817,
-      "required_pages_per_second_upper_estimate": 0.0017193745216825124
+      "required_token_positions_per_second": 28672.78955977483,
+      "required_pages_per_second_upper_estimate": 0.0017188899532938346
     },
     {
       "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": 28680.872641189817,
-      "required_pages_per_second_upper_estimate": 0.002229811694026107
+      "required_token_positions_per_second": 28672.78955977483,
+      "required_pages_per_second_upper_estimate": 0.0022291832699998094
     },
     {
       "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": 28680.872641189817,
-      "required_pages_per_second_upper_estimate": 0.36788743911942917
+      "required_token_positions_per_second": 28672.78955977483,
+      "required_pages_per_second_upper_estimate": 0.3677837580299743
     }
   ],
   "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": 0.13002529569507715,
-    "producer_p95_seconds": 0.30773573300029966,
-    "gpu_update_seconds": 0.963987405330671,
-    "producer_batches_per_second": 7.690811196808228,
-    "gpu_batches_per_second": 1.0373579514319233,
-    "producer_headroom": 7.413845130499236,
+    "producer_batch_seconds": 0.10190634278896482,
+    "producer_p95_seconds": 0.30179922900060774,
+    "gpu_update_seconds": 0.9642591608452179,
+    "producer_batches_per_second": 9.81293188070613,
+    "gpu_batches_per_second": 1.0370655946099112,
+    "producer_headroom": 9.462209460720977,
     "sustainable": true,
-    "buffer_seconds": 60.73120653583227,
+    "buffer_seconds": 60.74832713324873,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343015.0336850077,
-    "upload_bytes_per_second": 5434171.154848103,
+    "generation_bytes_per_second": 357976.83309128886,
+    "upload_bytes_per_second": 5345112.068085469,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.60304124748872,
+    "estimated_full_shards_per_hour": 76.81349510721206,
     "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": 1.0769242890019086,
-      "bytes": 7547252,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9740304760016443,
       "bytes": 5265802,
@@ -562,6 +556,12 @@
       "bytes": 7360891,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1906762940016051,
+      "bytes": 1163380,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -316,3 +316,4 @@
 {"step": 636000, "loss": 0.5701490044593811, "lr": 0.0001, "grad_norm": 0.17443959416940916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93093666340, "tokens_per_second": 151694.1968681244, "data_wait_s": 7.801998435752466e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 636100, "loss": 0.568419873714447, "lr": 0.0001, "grad_norm": 0.17273770364073185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93108323872, "tokens_per_second": 152065.60477108424, "data_wait_s": 1.0443000064697117e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 636200, "loss": 0.5601911544799805, "lr": 0.0001, "grad_norm": 0.1915028598132327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93122970603, "tokens_per_second": 152335.52874043494, "data_wait_s": 1.815399809856899e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 636300, "loss": 0.5318170189857483, "lr": 0.0001, "grad_norm": 0.17127548365267142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93137681205, "tokens_per_second": 152640.46434894545, "data_wait_s": 1.4750999980606139e-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 sha256:392d3956ea3dc7a8fbf998e4b407455b0fff01ed7307ef7e1865ff7773eb2c30
-size 679170
+oid sha256:1909a602d2ba9939ca09533c711ce5d49b702d32226bbd46b6079e2bcc5e671c
+size 681288