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": 76652.91397693462,
-      "required_pages_per_second_upper_estimate": 0.005303026448188988
+      "required_token_positions_per_second": 76852.6698112814,
+      "required_pages_per_second_upper_estimate": 0.005316846020306492
     },
     {
       "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": 9581.614247116828,
-      "required_pages_per_second_upper_estimate": 0.004011621798409701
+      "required_token_positions_per_second": 9606.583726410176,
+      "required_pages_per_second_upper_estimate": 0.004022075997967805
     },
     {
       "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": 19163.228494233656,
-      "required_pages_per_second_upper_estimate": 0.00399119512702893
+      "required_token_positions_per_second": 19213.16745282035,
+      "required_pages_per_second_upper_estimate": 0.004001596095123638
     },
     {
       "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": 28744.842741350483,
-      "required_pages_per_second_upper_estimate": 0.0017232094315104572
+      "required_token_positions_per_second": 28819.751179230527,
+      "required_pages_per_second_upper_estimate": 0.0017277000779828038
     },
     {
       "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": 28744.842741350483,
-      "required_pages_per_second_upper_estimate": 0.002234785087938865
+      "required_token_positions_per_second": 28819.751179230527,
+      "required_pages_per_second_upper_estimate": 0.002240608889496641
     },
     {
       "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": 28744.842741350483,
-      "required_pages_per_second_upper_estimate": 0.36870797887854806
+      "required_token_positions_per_second": 28819.751179230527,
+      "required_pages_per_second_upper_estimate": 0.3696688238892591
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.459840798960812,
-    "producer_p95_seconds": 0.3099871809972683,
-    "gpu_update_seconds": 0.9618421032523988,
-    "producer_batches_per_second": 2.1746656718148683,
-    "gpu_batches_per_second": 1.0396716848000032,
-    "producer_headroom": 2.091685003649204,
+    "producer_batch_seconds": 0.103063767156101,
+    "producer_p95_seconds": 0.3022905619982339,
+    "gpu_update_seconds": 0.9593420785646835,
+    "producer_batches_per_second": 9.702730916922473,
+    "gpu_batches_per_second": 1.042381046702493,
+    "producer_headroom": 9.308238045594221,
     "sustainable": true,
-    "buffer_seconds": 60.596052504901124,
+    "buffer_seconds": 60.43855094957506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277935.4490668699,
-    "upload_bytes_per_second": 5388973.568098741,
+    "generation_bytes_per_second": 284002.116762901,
+    "upload_bytes_per_second": 5436700.3416867545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.63847736363004,
+    "estimated_full_shards_per_hour": 60.940243026402214,
     "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": 1.175540739997814,
-      "bytes": 6085764,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1963462499988964,
       "bytes": 8278514,
@@ -562,6 +556,12 @@
       "bytes": 741591,
       "pages": 21,
       "shards": 1
+    },
+    {
+      "seconds": 1.6113828489978914,
+      "bytes": 12223061,
+      "pages": 49,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -305,3 +305,4 @@
 {"step": 470700, "loss": 0.5705721378326416, "lr": 0.0003, "grad_norm": 0.15167168461419328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68896108283, "tokens_per_second": 151601.6950746229, "data_wait_s": 1.0528001439524814e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 470800, "loss": 0.5753962397575378, "lr": 0.0003, "grad_norm": 0.18389009604001882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68910712220, "tokens_per_second": 152375.92854890318, "data_wait_s": 1.5951998648233712e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 470900, "loss": 0.6000497341156006, "lr": 0.0003, "grad_norm": 0.1599101394922151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68925446745, "tokens_per_second": 147862.43837226127, "data_wait_s": 1.0037998436018825e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 471000, "loss": 0.5956321358680725, "lr": 0.0003, "grad_norm": 0.16470233697343703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68940085663, "tokens_per_second": 152234.74635804413, "data_wait_s": 1.0285999451298267e-05, "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:e96936fc431d263628442b2d20acf2d454bec2ced025fc7c7fb4ed4c3698176b
-size 655089
+oid sha256:e5fce9edbd4e16152cdad01e39ce5e764ce6faaf0ab7313eb3f0272e3acdd97a
+size 657207