Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76019.62440150269,
-      "required_pages_per_second_upper_estimate": 0.006589188373145682
+      "required_token_positions_per_second": 75985.8752677039,
+      "required_pages_per_second_upper_estimate": 0.006586263083764401
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9502.453050187836,
-      "required_pages_per_second_upper_estimate": 0.003790670596053868
+      "required_token_positions_per_second": 9498.234408462988,
+      "required_pages_per_second_upper_estimate": 0.003788987716795511
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19004.906100375672,
-      "required_pages_per_second_upper_estimate": 0.004396450551398283
+      "required_token_positions_per_second": 18996.468816925975,
+      "required_pages_per_second_upper_estimate": 0.004394498734363312
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28507.35915056351,
-      "required_pages_per_second_upper_estimate": 0.001735948605714967
+      "required_token_positions_per_second": 28494.703225388963,
+      "required_pages_per_second_upper_estimate": 0.001735177926272345
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28507.35915056351,
-      "required_pages_per_second_upper_estimate": 0.0021945754208311913
+      "required_token_positions_per_second": 28494.703225388963,
+      "required_pages_per_second_upper_estimate": 0.0021936011326774115
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28507.35915056351,
-      "required_pages_per_second_upper_estimate": 0.3539659926563382
+      "required_token_positions_per_second": 28494.703225388963,
+      "required_pages_per_second_upper_estimate": 0.3538088484223272
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05822909157821954,
-    "producer_p95_seconds": 0.18585316099961346,
-    "gpu_update_seconds": 0.9698548313077775,
-    "producer_batches_per_second": 17.17354629612748,
-    "gpu_batches_per_second": 1.0310821452026733,
-    "producer_headroom": 16.655846845987025,
+    "producer_batch_seconds": 0.0693448702969448,
+    "producer_p95_seconds": 0.17827066800055036,
+    "gpu_update_seconds": 0.9702855924242599,
+    "producer_batches_per_second": 14.42067734380142,
+    "gpu_batches_per_second": 1.0306243932794041,
+    "producer_headroom": 13.992175459689465,
     "sustainable": true,
-    "buffer_seconds": 60.130999541082204,
+    "buffer_seconds": 60.15770673030411,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327174.9280786231,
-    "upload_bytes_per_second": 5913785.881491825,
+    "generation_bytes_per_second": 326186.28378404246,
+    "upload_bytes_per_second": 5794163.95626758,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.2041233231451,
+    "estimated_full_shards_per_hour": 69.9919832719894,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 12141179,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.1883426349995716,
+      "bytes": 1212298,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 47700, "loss": 0.644190676510334, "lr": 0.0003, "grad_norm": 0.13149846331555648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6980288879, "tokens_per_second": 145472.13143954135, "data_wait_s": 8.947001333581284e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47800, "loss": 0.6294682342559099, "lr": 0.0003, "grad_norm": 0.13426431920348905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6995001760, "tokens_per_second": 151139.69358223415, "data_wait_s": 8.867000360623933e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47900, "loss": 0.6194884553551674, "lr": 0.0003, "grad_norm": 0.13019894367045043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7009648224, "tokens_per_second": 150052.20525703835, "data_wait_s": 9.747000149218366e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 48000, "loss": 0.6674314849078655, "lr": 0.0003, "grad_norm": 0.13019126365460446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7024387599, "tokens_per_second": 150752.32271111163, "data_wait_s": 1.6042999050114304e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cbb0ca1051efe74048e12b3c399496d0559248bc55e2f423edb6e326dd03d9f
-size 168270
+oid sha256:9e1da2ab3af0af5c619ca816bcaaf1990b5537a0d276f6b4d219b17d5a61dfca
+size 170388