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": 76253.67952125106,
-      "required_pages_per_second_upper_estimate": 0.005275406482193236
+      "required_token_positions_per_second": 76787.61513387342,
+      "required_pages_per_second_upper_estimate": 0.005312345386775759
     },
     {
       "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": 9531.709940156383,
-      "required_pages_per_second_upper_estimate": 0.003990727907205795
+      "required_token_positions_per_second": 9598.451891734177,
+      "required_pages_per_second_upper_estimate": 0.004018671368600983
     },
     {
       "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": 19063.419880312766,
-      "required_pages_per_second_upper_estimate": 0.0039704076248793605
+      "required_token_positions_per_second": 19196.903783468355,
+      "required_pages_per_second_upper_estimate": 0.00399820880169942
     },
     {
       "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": 28595.12982046915,
-      "required_pages_per_second_upper_estimate": 0.001714234370502015
+      "required_token_positions_per_second": 28795.355675202532,
+      "required_pages_per_second_upper_estimate": 0.0017262376047661033
     },
     {
       "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": 28595.12982046915,
-      "required_pages_per_second_upper_estimate": 0.002223145566857886
+      "required_token_positions_per_second": 28795.355675202532,
+      "required_pages_per_second_upper_estimate": 0.002238712246362947
     },
     {
       "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": 28595.12982046915,
-      "required_pages_per_second_upper_estimate": 0.36678762227869255
+      "required_token_positions_per_second": 28795.355675202532,
+      "required_pages_per_second_upper_estimate": 0.3693559045574394
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14687321671877385,
-    "producer_p95_seconds": 0.1997672710003826,
-    "gpu_update_seconds": 0.9668779324865604,
-    "producer_batches_per_second": 6.8085933047599445,
-    "gpu_batches_per_second": 1.0342567209371076,
-    "producer_headroom": 6.583078617648133,
+    "producer_batch_seconds": 0.0602029283751051,
+    "producer_p95_seconds": 0.19551041100021394,
+    "gpu_update_seconds": 0.9601548357956005,
+    "producer_batches_per_second": 16.610487678760762,
+    "gpu_batches_per_second": 1.0414986861690732,
+    "producer_headroom": 15.948640069685384,
     "sustainable": true,
-    "buffer_seconds": 60.913309746653304,
+    "buffer_seconds": 60.48975465512283,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279231.65536078595,
+    "generation_bytes_per_second": 295988.0364706684,
     "upload_bytes_per_second": 5500821.607422729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.916613060166206,
+    "estimated_full_shards_per_hour": 63.512142377758394,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 771000, "loss": 0.492196261882782, "lr": 3e-05, "grad_norm": 0.23602603074079667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112852090321, "tokens_per_second": 150908.68533219758, "data_wait_s": 1.5677000192226842e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 771100, "loss": 0.4970531463623047, "lr": 3e-05, "grad_norm": 0.23643469945190743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112866569411, "tokens_per_second": 150266.1260719092, "data_wait_s": 9.981000403058715e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 771200, "loss": 0.5060374736785889, "lr": 3e-05, "grad_norm": 0.25702834694377064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112881244233, "tokens_per_second": 152751.98276192133, "data_wait_s": 1.2203001460875385e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 771300, "loss": 0.5389420986175537, "lr": 3e-05, "grad_norm": 0.24801907482809776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112895902196, "tokens_per_second": 152771.32929801662, "data_wait_s": 8.81000050867442e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b37f84bc64c4d77b11df2e81252627fa8705f2e169f57973e133a94b4d002b4c
-size 238947
+oid sha256:3503f744a4dc117428416b0716f29585697e8294d4942558e89a9a8a63202a02
+size 241065