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": 75682.71760821706,
-      "required_pages_per_second_upper_estimate": 0.006559986146186145
+      "required_token_positions_per_second": 75466.82932902394,
+      "required_pages_per_second_upper_estimate": 0.006541273497309526
     },
     {
       "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": 9460.339701027133,
-      "required_pages_per_second_upper_estimate": 0.0037738709514229824
+      "required_token_positions_per_second": 9433.353666127992,
+      "required_pages_per_second_upper_estimate": 0.003763105818624538
     },
     {
       "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": 18920.679402054266,
-      "required_pages_per_second_upper_estimate": 0.004376966187080887
+      "required_token_positions_per_second": 18866.707332255985,
+      "required_pages_per_second_upper_estimate": 0.004364480698609041
     },
     {
       "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": 28381.0191030814,
-      "required_pages_per_second_upper_estimate": 0.0017282551596783076
+      "required_token_positions_per_second": 28300.06099838398,
+      "required_pages_per_second_upper_estimate": 0.0017233252358565833
     },
     {
       "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": 28381.0191030814,
-      "required_pages_per_second_upper_estimate": 0.002184849414244383
+      "required_token_positions_per_second": 28300.06099838398,
+      "required_pages_per_second_upper_estimate": 0.0021786170352383985
     },
     {
       "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": 28381.0191030814,
-      "required_pages_per_second_upper_estimate": 0.35239727209954924
+      "required_token_positions_per_second": 28300.06099838398,
+      "required_pages_per_second_upper_estimate": 0.35139204338855407
     }
   ],
   "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": 0.10513727747652979,
-    "producer_p95_seconds": 0.20851804500125581,
-    "gpu_update_seconds": 0.9741722064165831,
-    "producer_batches_per_second": 9.511374309870579,
-    "gpu_batches_per_second": 1.0265125543649232,
-    "producer_headroom": 9.265716497500627,
+    "producer_batch_seconds": 0.07672668312496,
+    "producer_p95_seconds": 0.21722265499920468,
+    "gpu_update_seconds": 0.9769590249850977,
+    "producer_batches_per_second": 13.033275508226543,
+    "gpu_batches_per_second": 1.0235843821753463,
+    "producer_headroom": 12.732976132879157,
     "sustainable": true,
-    "buffer_seconds": 60.39867679782815,
+    "buffer_seconds": 60.57145954907605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485146.8528922552,
-    "upload_bytes_per_second": 6009279.239589546,
+    "generation_bytes_per_second": 499029.3521982978,
+    "upload_bytes_per_second": 5988272.676300859,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.10122098994962,
+    "estimated_full_shards_per_hour": 107.08008217298222,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 5923485,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.4506824640011473,
+      "bytes": 7595330,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 125500, "loss": 0.6334497854113579, "lr": 0.0003, "grad_norm": 0.1416959142518137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18365841451, "tokens_per_second": 148438.85682220917, "data_wait_s": 1.3680999472853728e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125600, "loss": 0.6110418401658535, "lr": 0.0003, "grad_norm": 0.143462033952661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18380539739, "tokens_per_second": 150167.48174788724, "data_wait_s": 9.536001016385853e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 125700, "loss": 0.6051217522472143, "lr": 0.0003, "grad_norm": 0.13412204190661894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18395297106, "tokens_per_second": 150093.03860580086, "data_wait_s": 1.2066999261151068e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 125800, "loss": 0.5929410122334957, "lr": 0.0003, "grad_norm": 0.13166707405003372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18410100520, "tokens_per_second": 150139.31829533188, "data_wait_s": 1.6150999726960436e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39256b41b86e1cd2a66b8aad87e24ae2eeb0de3a55abef8d2fd58a89e039cc13
-size 228357
+oid sha256:6465c5b78153e393a332afd75429db3706b7ddec8d8ac3df10f8106c0bbab456
+size 230475