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": 76368.21579001032,
-      "required_pages_per_second_upper_estimate": 0.00528333036702676
+      "required_token_positions_per_second": 77050.42122520902,
+      "required_pages_per_second_upper_estimate": 0.005330526921968501
     },
     {
       "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": 9546.02697375129,
-      "required_pages_per_second_upper_estimate": 0.003996722150198324
+      "required_token_positions_per_second": 9631.302653151128,
+      "required_pages_per_second_upper_estimate": 0.004032425296404353
     },
     {
       "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": 19092.05394750258,
-      "required_pages_per_second_upper_estimate": 0.003976371345943867
+      "required_token_positions_per_second": 19262.605306302255,
+      "required_pages_per_second_upper_estimate": 0.004011892696240048
     },
     {
       "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": 28638.08092125387,
-      "required_pages_per_second_upper_estimate": 0.0017168092234115255
+      "required_token_positions_per_second": 28893.907959453383,
+      "required_pages_per_second_upper_estimate": 0.001732145663726317
     },
     {
       "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": 28638.08092125387,
-      "required_pages_per_second_upper_estimate": 0.0022264848260219712
+      "required_token_positions_per_second": 28893.907959453383,
+      "required_pages_per_second_upper_estimate": 0.002246374252977764
     },
     {
       "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": 28638.08092125387,
-      "required_pages_per_second_upper_estimate": 0.36733855288225997
+      "required_token_positions_per_second": 28893.907959453383,
+      "required_pages_per_second_upper_estimate": 0.37062002744261086
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.8098722178517619,
-    "producer_p95_seconds": 0.27143339500253205,
-    "gpu_update_seconds": 0.965427818855031,
-    "producer_batches_per_second": 1.2347626921350188,
-    "gpu_batches_per_second": 1.0358102185059992,
-    "producer_headroom": 1.1920742526714774,
+    "producer_batch_seconds": 0.1993020477032701,
+    "producer_p95_seconds": 0.26631415899828426,
+    "gpu_update_seconds": 0.9568799083460169,
+    "producer_batches_per_second": 5.01750991283765,
+    "gpu_batches_per_second": 1.0450632219130997,
+    "producer_headroom": 4.8011544255213225,
     "sustainable": true,
-    "buffer_seconds": 60.82195258786695,
+    "buffer_seconds": 60.28343422579906,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 30737.992186955864,
-    "upload_bytes_per_second": 4889308.79492777,
+    "generation_bytes_per_second": 31019.482101510057,
+    "upload_bytes_per_second": 4986663.792060345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 6.595657579484052,
+    "estimated_full_shards_per_hour": 6.65605876239754,
     "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": 1.240830022998125,
-      "bytes": 6131822,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2179747789996327,
       "bytes": 1206466,
@@ -562,6 +556,12 @@
       "bytes": 4403386,
       "pages": 18,
       "shards": 2
+    },
+    {
+      "seconds": 1.2122206399981224,
+      "bytes": 13761400,
+      "pages": 50,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 711500, "loss": 0.5800469517707825, "lr": 8e-05, "grad_norm": 0.21757968886907136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104143551438, "tokens_per_second": 130549.11382430035, "data_wait_s": 1.709700154606253e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711600, "loss": 0.5820253491401672, "lr": 8e-05, "grad_norm": 0.21575630043191904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104158248858, "tokens_per_second": 111436.34439939866, "data_wait_s": 1.4609002391807735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711700, "loss": 0.5247791409492493, "lr": 8e-05, "grad_norm": 0.21384825472213356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104172817531, "tokens_per_second": 151856.43157379143, "data_wait_s": 2.0504001440713182e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 711800, "loss": 0.5453563332557678, "lr": 8e-05, "grad_norm": 0.2102621463966983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104187343352, "tokens_per_second": 150597.08948802997, "data_wait_s": 1.7242000467376783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4d371e66ac0befa46cec66edc533fd7e390c1b423c63d62f160c17d2451f1ba
-size 569586
+oid sha256:5e772fcffcee567ed40526c1c54898b7b66aaa308eb7dc9b105e03b974c2ac2a
+size 571704