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": 75967.22746347329,
-      "required_pages_per_second_upper_estimate": 0.006584646739356119
+      "required_token_positions_per_second": 75436.00239654537,
+      "required_pages_per_second_upper_estimate": 0.00653860149693243
     },
     {
       "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": 9495.903432934161,
-      "required_pages_per_second_upper_estimate": 0.0037880578558058724
+      "required_token_positions_per_second": 9429.500299568172,
+      "required_pages_per_second_upper_estimate": 0.0037615686530908616
     },
     {
       "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": 18991.806865868322,
-      "required_pages_per_second_upper_estimate": 0.004393420274033661
+      "required_token_positions_per_second": 18859.000599136343,
+      "required_pages_per_second_upper_estimate": 0.004362697881535683
     },
     {
       "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": 28487.71029880248,
-      "required_pages_per_second_upper_estimate": 0.0017347520937322784
+      "required_token_positions_per_second": 28288.500898704515,
+      "required_pages_per_second_upper_estimate": 0.0017226212864372597
     },
     {
       "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": 28487.71029880248,
-      "required_pages_per_second_upper_estimate": 0.0021930627978311245
+      "required_token_positions_per_second": 28288.500898704515,
+      "required_pages_per_second_upper_estimate": 0.0021777271067646695
     },
     {
       "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": 28487.71029880248,
-      "required_pages_per_second_upper_estimate": 0.3537220196779428
+      "required_token_positions_per_second": 28288.500898704515,
+      "required_pages_per_second_upper_estimate": 0.35124850563969995
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07762978306240598,
-    "producer_p95_seconds": 0.1995838780003396,
-    "gpu_update_seconds": 0.9705237700750635,
-    "producer_batches_per_second": 12.881653929086827,
-    "gpu_batches_per_second": 1.0303714662471963,
-    "producer_headroom": 12.501951336059603,
+    "producer_batch_seconds": 0.38050877037498765,
+    "producer_p95_seconds": 0.2422439990004932,
+    "gpu_update_seconds": 0.977358259421451,
+    "producer_batches_per_second": 2.628060317806893,
+    "gpu_batches_per_second": 1.023166265144116,
+    "producer_headroom": 2.5685564578663302,
     "sustainable": true,
-    "buffer_seconds": 60.172473744653935,
+    "buffer_seconds": 60.59621208412997,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 191797.30140295514,
-    "upload_bytes_per_second": 6593767.810258815,
+    "generation_bytes_per_second": 187927.3360595079,
+    "upload_bytes_per_second": 6523426.966652929,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 41.15523606840602,
+    "estimated_full_shards_per_hour": 40.32483159388474,
     "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": 0.8969114139999874,
-      "bytes": 7503017,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8951566169998841,
       "bytes": 8341517,
@@ -562,6 +556,12 @@
       "bytes": 16761254,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.7977973630004271,
+      "bytes": 2044807,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 9700, "loss": 0.7019376754760742, "lr": 0.0003, "grad_norm": 0.15032665239691817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1418274502, "tokens_per_second": 148051.27805517896, "data_wait_s": 1.4580999049940147e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9800, "loss": 0.6842157915234566, "lr": 0.0003, "grad_norm": 0.14682688827643217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1432900392, "tokens_per_second": 92317.45613918074, "data_wait_s": 1.2158001482021064e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9900, "loss": 0.801698736846447, "lr": 0.0003, "grad_norm": 0.1893745328406307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1447622497, "tokens_per_second": 151120.41772944975, "data_wait_s": 8.74599936651066e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 10000, "loss": 0.7166141532361507, "lr": 0.0003, "grad_norm": 0.17501825246700883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1462271010, "tokens_per_second": 148867.69857601103, "data_wait_s": 8.556000466342084e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b642558b264e1fdf879891ddee550aae4c771ae0ea1eed27abfb269bbea9b23
-size 207116
+oid sha256:172a1e6f2cbb6b66a5a0f214ba322859d17d29c80ff06e7a58887bc2ee13269c
+size 209199