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": 75399.26931698152,
-      "required_pages_per_second_upper_estimate": 0.0052162964017570595
+      "required_token_positions_per_second": 75446.6721857848,
+      "required_pages_per_second_upper_estimate": 0.005219575842210679
     },
     {
       "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": 9424.90866462269,
-      "required_pages_per_second_upper_estimate": 0.003946012443404083
+      "required_token_positions_per_second": 9430.8340232231,
+      "required_pages_per_second_upper_estimate": 0.003948493267314517
     },
     {
       "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": 18849.81732924538,
-      "required_pages_per_second_upper_estimate": 0.003925919846570104
+      "required_token_positions_per_second": 18861.6680464462,
+      "required_pages_per_second_upper_estimate": 0.003928388038438606
     },
     {
       "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": 28274.72599386807,
-      "required_pages_per_second_upper_estimate": 0.0016950266503255962
+      "required_token_positions_per_second": 28292.5020696693,
+      "required_pages_per_second_upper_estimate": 0.0016960922989273839
     },
     {
       "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": 28274.72599386807,
-      "required_pages_per_second_upper_estimate": 0.0021982355786471447
+      "required_token_positions_per_second": 28292.5020696693,
+      "required_pages_per_second_upper_estimate": 0.002199617590351996
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28274.72599386807,
-      "required_pages_per_second_upper_estimate": 0.3626778260138796
+      "required_token_positions_per_second": 28292.5020696693,
+      "required_pages_per_second_upper_estimate": 0.36290583842779467
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09721066243767496,
-    "producer_p95_seconds": 0.23875339799997164,
-    "gpu_update_seconds": 0.9778344096418835,
-    "producer_batches_per_second": 10.286937409166756,
-    "gpu_batches_per_second": 1.0226680408661772,
-    "producer_headroom": 10.05892136851558,
+    "producer_batch_seconds": 0.08674492598419192,
+    "producer_p95_seconds": 0.23453101300037815,
+    "gpu_update_seconds": 0.977220039850762,
+    "producer_batches_per_second": 11.52805179846757,
+    "gpu_batches_per_second": 1.0233109834226455,
+    "producer_headroom": 11.265443237900127,
     "sustainable": true,
-    "buffer_seconds": 60.62573339779678,
+    "buffer_seconds": 60.58764247074724,
     "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": 298240.40982654464,
+    "generation_bytes_per_second": 298503.88936027576,
     "upload_bytes_per_second": 3854490.630778565,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.9954492673612,
+    "estimated_full_shards_per_hour": 64.05198584181026,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 818400, "loss": 0.49868494272232056, "lr": 3e-05, "grad_norm": 0.2709542075351307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119790571490, "tokens_per_second": 152320.76112192284, "data_wait_s": 8.18000262370333e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 818500, "loss": 0.4756673276424408, "lr": 3e-05, "grad_norm": 0.27991657973072104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119805262667, "tokens_per_second": 151835.13705665397, "data_wait_s": 6.9779998739250004e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 818600, "loss": 0.5501465797424316, "lr": 3e-05, "grad_norm": 0.28155200234403827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119819942529, "tokens_per_second": 151043.82768929985, "data_wait_s": 9.424999007023871e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 818700, "loss": 0.565858781337738, "lr": 3e-05, "grad_norm": 0.27960762824885566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119834757197, "tokens_per_second": 153248.86429755297, "data_wait_s": 8.236998837674037e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96cde5bf84eb2bd8bfeb533c9a2c90c313968312a8974d5ee401a3c94accb185
-size 396462
+oid sha256:afbe85f4058009e3b282736faf57e04141cfbe03ef59f98204f8c858d0dc9ca3
+size 398580