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": 74957.93170884748,
-      "required_pages_per_second_upper_estimate": 0.006497163541381857
+      "required_token_positions_per_second": 73791.10898194685,
+      "required_pages_per_second_upper_estimate": 0.0063960263046992745
     },
     {
       "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": 9369.741463605935,
-      "required_pages_per_second_upper_estimate": 0.0037377299599513063
+      "required_token_positions_per_second": 9223.888622743356,
+      "required_pages_per_second_upper_estimate": 0.003679547081036922
     },
     {
       "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": 18739.48292721187,
-      "required_pages_per_second_upper_estimate": 0.00433504957157514
+      "required_token_positions_per_second": 18447.77724548671,
+      "required_pages_per_second_upper_estimate": 0.004267568595952681
     },
     {
       "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": 28109.224390817802,
-      "required_pages_per_second_upper_estimate": 0.001711704287698103
+      "required_token_positions_per_second": 27671.66586823007,
+      "required_pages_per_second_upper_estimate": 0.0016850592693646576
     },
     {
       "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": 28109.224390817802,
-      "required_pages_per_second_upper_estimate": 0.0021639259049184138
+      "required_token_positions_per_second": 27671.66586823007,
+      "required_pages_per_second_upper_estimate": 0.0021302414386100968
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28109.224390817802,
-      "required_pages_per_second_upper_estimate": 0.349022491411622
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27671.66586823007,
+      "required_pages_per_second_upper_estimate": 0.3435894789752545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12849218111722394,
-    "producer_p95_seconds": 0.3934429660002934,
-    "gpu_update_seconds": 0.9835917069640503,
-    "producer_batches_per_second": 7.782574716259941,
-    "gpu_batches_per_second": 1.0166820164502968,
-    "producer_headroom": 7.654875949741375,
+    "producer_batch_seconds": 0.12881192678094067,
+    "producer_p95_seconds": 0.42765358700125944,
+    "gpu_update_seconds": 0.9991447617088627,
+    "producer_batches_per_second": 7.763256283718306,
+    "gpu_batches_per_second": 1.0008559703497566,
+    "producer_headroom": 7.756616849680558,
     "sustainable": true,
-    "buffer_seconds": 60.98268583177112,
+    "buffer_seconds": 60.947830464240624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401419.197211891,
-    "upload_bytes_per_second": 5776158.800846744,
+    "generation_bytes_per_second": 430297.26910242974,
+    "upload_bytes_per_second": 5798552.363700355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.13521516101406,
+    "estimated_full_shards_per_hour": 92.33177714161558,
     "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": 1.256377692996466,
-      "bytes": 6950847,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3214142249998986,
       "bytes": 10934190,
@@ -562,6 +556,12 @@
       "bytes": 919043,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.127943243001937,
+      "bytes": 7825410,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 396500, "loss": 0.5753758829087019, "lr": 0.0003, "grad_norm": 0.1413881660238825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58036253337, "tokens_per_second": 149687.7418879604, "data_wait_s": 1.720000000204891e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396600, "loss": 0.6008331198245287, "lr": 0.0003, "grad_norm": 0.1821427877255445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58050988572, "tokens_per_second": 150142.8073788414, "data_wait_s": 1.4903998817317188e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396700, "loss": 0.6057343687862158, "lr": 0.0003, "grad_norm": 0.18002764965165677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58065761046, "tokens_per_second": 149631.85785415737, "data_wait_s": 9.059993317350745e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 396800, "loss": 0.600704200565815, "lr": 0.0003, "grad_norm": 0.156986707827914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58080489915, "tokens_per_second": 150732.682964153, "data_wait_s": 1.0575000487733632e-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 sha256:4b757a354fad7f45652b92cfecd8c8d9e9ccdce06cc400054bc6f8cdd9752c5d
-size 889635
+oid sha256:b7e4b9e2fcb76875072375a31ff702ecf09ff95d02d842e0170738463ab921a6
+size 891753