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": 76851.69119434712,
-      "required_pages_per_second_upper_estimate": 0.005316778317316274
+      "required_token_positions_per_second": 76953.730982196,
+      "required_pages_per_second_upper_estimate": 0.005323837666604586
     },
     {
       "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": 9606.46139929339,
-      "required_pages_per_second_upper_estimate": 0.004022024782158488
+      "required_token_positions_per_second": 9619.2163727745,
+      "required_pages_per_second_upper_estimate": 0.004027365023200893
     },
     {
       "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": 19212.92279858678,
-      "required_pages_per_second_upper_estimate": 0.004001545140098751
+      "required_token_positions_per_second": 19238.432745549,
+      "required_pages_per_second_upper_estimate": 0.00400685818930844
     },
     {
       "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": 28819.384197880172,
-      "required_pages_per_second_upper_estimate": 0.001727678078011198
+      "required_token_positions_per_second": 28857.6491183235,
+      "required_pages_per_second_upper_estimate": 0.0017299720015646251
     },
     {
       "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": 28819.384197880172,
-      "required_pages_per_second_upper_estimate": 0.002240580358310832
+      "required_token_positions_per_second": 28857.6491183235,
+      "required_pages_per_second_upper_estimate": 0.002243555287565703
     },
     {
       "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": 28819.384197880172,
-      "required_pages_per_second_upper_estimate": 0.3696641166465306
+      "required_token_positions_per_second": 28857.6491183235,
+      "required_pages_per_second_upper_estimate": 0.3701549379603071
     }
   ],
   "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": 0.3907976188905451,
-    "producer_p95_seconds": 0.2491051739998511,
-    "gpu_update_seconds": 0.9593542946706567,
-    "producer_batches_per_second": 2.5588692245335323,
-    "gpu_batches_per_second": 1.0423677733608279,
-    "producer_headroom": 2.454862180056817,
+    "producer_batch_seconds": 0.09608104714847343,
+    "producer_p95_seconds": 0.24937971600047604,
+    "gpu_update_seconds": 0.9580822015901698,
+    "producer_batches_per_second": 10.407879906374317,
+    "gpu_batches_per_second": 1.043751776559733,
+    "producer_headroom": 9.971604494585195,
     "sustainable": true,
-    "buffer_seconds": 60.439320564251375,
+    "buffer_seconds": 60.35917870018069,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332940.3113819717,
-    "upload_bytes_per_second": 5300138.259366258,
+    "generation_bytes_per_second": 320174.74108804803,
+    "upload_bytes_per_second": 5203145.385523799,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.44124036878932,
+    "estimated_full_shards_per_hour": 68.70204615098076,
     "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": 1.5761228700002903,
-      "bytes": 18281686,
-      "pages": 39,
-      "shards": 5
-    },
     {
       "seconds": 1.1403170100002171,
       "bytes": 5965661,
@@ -562,6 +556,12 @@
       "bytes": 5975341,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.473726465999789,
+      "bytes": 9735362,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 900400, "loss": 0.554978609085083, "lr": 3e-05, "grad_norm": 0.32366369021997676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131793093122, "tokens_per_second": 151810.67255527276, "data_wait_s": 1.5450999853783287e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 900500, "loss": 0.577768862247467, "lr": 3e-05, "grad_norm": 0.3444207151321536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131807727087, "tokens_per_second": 151676.22166932866, "data_wait_s": 1.2712000170722604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 900600, "loss": 0.519142210483551, "lr": 3e-05, "grad_norm": 0.3188655176671339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131822449563, "tokens_per_second": 153689.71074172438, "data_wait_s": 1.2581998817040585e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 900700, "loss": 0.5623064041137695, "lr": 3e-05, "grad_norm": 0.3314885858898911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131837058499, "tokens_per_second": 153046.05133180146, "data_wait_s": 1.0079000276164152e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4eb6a17d188453ba16962d6052340b72f821b7a8cb77d951c2131715b9377e17
-size 332922
+oid sha256:f31e5c3adbdf986495ab6b8afd26ad353a439b91c194cf838c64bc54bba89612
+size 335040