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": 75535.20680659584,
-      "required_pages_per_second_upper_estimate": 0.006547200283764327
+      "required_token_positions_per_second": 75247.27283888582,
+      "required_pages_per_second_upper_estimate": 0.006522242897205191
     },
     {
       "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": 9441.90085082448,
-      "required_pages_per_second_upper_estimate": 0.0037665154183917668
+      "required_token_positions_per_second": 9405.909104860728,
+      "required_pages_per_second_upper_estimate": 0.003752157772802269
     },
     {
       "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": 18883.80170164896,
-      "required_pages_per_second_upper_estimate": 0.00436843518011748
+      "required_token_positions_per_second": 18811.818209721456,
+      "required_pages_per_second_upper_estimate": 0.0043517830661262405
     },
     {
       "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": 28325.702552473445,
-      "required_pages_per_second_upper_estimate": 0.0017248866719697946
+      "required_token_positions_per_second": 28217.727314582185,
+      "required_pages_per_second_upper_estimate": 0.0017183115464845084
     },
     {
       "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": 28325.702552473445,
-      "required_pages_per_second_upper_estimate": 0.0021805909930525703
+      "required_token_positions_per_second": 28217.727314582185,
+      "required_pages_per_second_upper_estimate": 0.0021722787603451125
     },
     {
       "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": 28325.702552473445,
-      "required_pages_per_second_upper_estimate": 0.3517104256735841
+      "required_token_positions_per_second": 28217.727314582185,
+      "required_pages_per_second_upper_estimate": 0.3503697345888497
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10876319032018955,
-    "producer_p95_seconds": 0.30206424499920104,
-    "gpu_update_seconds": 0.9760746427659474,
-    "producer_batches_per_second": 9.194287121001924,
-    "gpu_batches_per_second": 1.0245118110703648,
-    "producer_headroom": 8.974310517119504,
+    "producer_batch_seconds": 0.27758744189839035,
+    "producer_p95_seconds": 0.3631653060001554,
+    "gpu_update_seconds": 0.9798095959950763,
+    "producer_batches_per_second": 3.6024684443975876,
+    "gpu_batches_per_second": 1.0206064566906172,
+    "producer_headroom": 3.5297331510902112,
     "sustainable": true,
-    "buffer_seconds": 60.516627851488735,
+    "buffer_seconds": 60.748194951694735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507573.62912412884,
+    "generation_bytes_per_second": 503829.6999343424,
     "upload_bytes_per_second": 4998444.908631304,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.91348510067843,
+    "estimated_full_shards_per_hour": 108.11012505076127,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -291,3 +291,4 @@
 {"step": 384100, "loss": 0.6103178188204765, "lr": 0.0003, "grad_norm": 0.1743308913183836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56221211483, "tokens_per_second": 147885.76171274134, "data_wait_s": 9.47399894357659e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 384200, "loss": 0.5748583097010851, "lr": 0.0003, "grad_norm": 0.14552985249548286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56235902421, "tokens_per_second": 150183.19092437477, "data_wait_s": 1.3942000805400312e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 384300, "loss": 0.620120232924819, "lr": 0.0003, "grad_norm": 0.160811714798311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56250704808, "tokens_per_second": 151232.42146370944, "data_wait_s": 1.9361999875400215e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 384400, "loss": 0.5866770055145025, "lr": 0.0003, "grad_norm": 0.14271838776488877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56265323151, "tokens_per_second": 148834.5039853025, "data_wait_s": 1.0653002391336486e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cca2530d7d3105943529d40d660c9911b493680cbec1dbd067f2dd584874b196
-size 624654
+oid sha256:ad70dab44be1140e724d0a6ddcedcd9b6c2aecc5c7e8647446d86797634e00bc
+size 626772