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": 75850.34625209444,
-      "required_pages_per_second_upper_estimate": 0.006574515772186523
+      "required_token_positions_per_second": 75655.97500746157,
+      "required_pages_per_second_upper_estimate": 0.006557668165331161
     },
     {
       "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": 9481.293281511806,
-      "required_pages_per_second_upper_estimate": 0.003782229647962265
+      "required_token_positions_per_second": 9456.996875932697,
+      "required_pages_per_second_upper_estimate": 0.0037725374485131228
     },
     {
       "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": 18962.58656302361,
-      "required_pages_per_second_upper_estimate": 0.004386660671230302
+      "required_token_positions_per_second": 18913.993751865393,
+      "required_pages_per_second_upper_estimate": 0.004375419579573116
     },
     {
       "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": 28443.87984453542,
-      "required_pages_per_second_upper_estimate": 0.0017320830490280357
+      "required_token_positions_per_second": 28370.990627798088,
+      "required_pages_per_second_upper_estimate": 0.0017276444781488974
     },
     {
       "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": 28443.87984453542,
-      "required_pages_per_second_upper_estimate": 0.002189688608131183
+      "required_token_positions_per_second": 28370.990627798088,
+      "required_pages_per_second_upper_estimate": 0.0021840773944564774
     },
     {
       "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": 28443.87984453542,
-      "required_pages_per_second_upper_estimate": 0.35317779212704
+      "required_token_positions_per_second": 28370.990627798088,
+      "required_pages_per_second_upper_estimate": 0.3522727519996783
     }
   ],
   "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": 0.11890119333605753,
-    "producer_p95_seconds": 0.2862573290003638,
-    "gpu_update_seconds": 0.9720192938204835,
-    "producer_batches_per_second": 8.410344521720992,
-    "gpu_batches_per_second": 1.0287861633584858,
-    "producer_headroom": 8.17501714279021,
+    "producer_batch_seconds": 0.09885962411726723,
+    "producer_p95_seconds": 0.333527533999586,
+    "gpu_update_seconds": 0.9745165532891299,
+    "producer_batches_per_second": 10.115353046596663,
+    "gpu_batches_per_second": 1.0261498346281137,
+    "producer_headroom": 9.85757898627208,
     "sustainable": true,
-    "buffer_seconds": 60.265196216869974,
+    "buffer_seconds": 60.42002630392605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367694.9498763091,
-    "upload_bytes_per_second": 3963258.320715347,
+    "generation_bytes_per_second": 383078.0327171926,
+    "upload_bytes_per_second": 3923941.0809006104,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.89877674309687,
+    "estimated_full_shards_per_hour": 82.19962822090945,
     "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": 1.0809610530013742,
-      "bytes": 6077185,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1773543239996798,
       "bytes": 6977422,
@@ -562,6 +556,12 @@
       "bytes": 11028707,
       "pages": 30,
       "shards": 4
+    },
+    {
+      "seconds": 0.9246157949964982,
+      "bytes": 1204930,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 138200, "loss": 0.6266972068697214, "lr": 0.0003, "grad_norm": 0.13924538543768078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20225549777, "tokens_per_second": 149880.93232815288, "data_wait_s": 2.473600034136325e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138300, "loss": 0.5591120338067412, "lr": 0.0003, "grad_norm": 0.12904902226672182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20239986214, "tokens_per_second": 145486.039391103, "data_wait_s": 1.326400160905905e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 138400, "loss": 0.6374600455164909, "lr": 0.0003, "grad_norm": 0.16114978434019975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20254527171, "tokens_per_second": 148251.54884230747, "data_wait_s": 1.8317001376999542e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 138500, "loss": 0.5999759268015623, "lr": 0.0003, "grad_norm": 0.14957302639950082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20269059703, "tokens_per_second": 147830.30864382494, "data_wait_s": 1.2942000466864556e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13b609ec8fab212bf9d27bb5a8348c3236484fba6e5e79152b0e21aabdd78816
-size 498909
+oid sha256:c4fcdd8eb7ab4424dc289f8d9a6477bbfc9dccfec980ebf30fa0abe97f48edb1
+size 501027