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": 74278.76964759517,
-      "required_pages_per_second_upper_estimate": 0.006438295495233014
+      "required_token_positions_per_second": 76011.44851677855,
+      "required_pages_per_second_upper_estimate": 0.006588479708179384
     },
     {
       "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": 9284.846205949396,
-      "required_pages_per_second_upper_estimate": 0.0037038639723748984
+      "required_token_positions_per_second": 9501.431064597318,
+      "required_pages_per_second_upper_estimate": 0.003790262910721764
     },
     {
       "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": 18569.69241189879,
-      "required_pages_per_second_upper_estimate": 0.004295771524068469
+      "required_token_positions_per_second": 19002.862129194637,
+      "required_pages_per_second_upper_estimate": 0.004395977714638214
     },
     {
       "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": 27854.53861784819,
-      "required_pages_per_second_upper_estimate": 0.0016961952603572349
+      "required_token_positions_per_second": 28504.293193791957,
+      "required_pages_per_second_upper_estimate": 0.0017357619050334114
     },
     {
       "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": 27854.53861784819,
-      "required_pages_per_second_upper_estimate": 0.0021443194890998704
+      "required_token_positions_per_second": 28504.293193791957,
+      "required_pages_per_second_upper_estimate": 0.0021943393949918058
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27854.53861784819,
-      "required_pages_per_second_upper_estimate": 0.34586014648978963
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28504.293193791957,
+      "required_pages_per_second_upper_estimate": 0.35392792373433274
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14240706428881822,
-    "producer_p95_seconds": 0.3709156859986251,
-    "gpu_update_seconds": 0.9925851000197201,
-    "producer_batches_per_second": 7.022123551201665,
-    "gpu_batches_per_second": 1.0074702914441618,
-    "producer_headroom": 6.970055207420337,
+    "producer_batch_seconds": 0.09074777631252573,
+    "producer_p95_seconds": 0.34983883099630475,
+    "gpu_update_seconds": 0.9699591500841547,
+    "producer_batches_per_second": 11.019553763566678,
+    "gpu_batches_per_second": 1.0309712526689798,
+    "producer_headroom": 10.688517002815784,
     "sustainable": true,
-    "buffer_seconds": 60.54769110120293,
+    "buffer_seconds": 60.13746730521759,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296575.8770452781,
-    "upload_bytes_per_second": 5252448.045226058,
+    "generation_bytes_per_second": 296807.9644553606,
+    "upload_bytes_per_second": 5272264.636593416,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.63827928084142,
+    "estimated_full_shards_per_hour": 63.68807983632673,
     "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": 0.9795328820000577,
-      "bytes": 930296,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.2613436630017532,
       "bytes": 7588531,
@@ -562,6 +556,12 @@
       "bytes": 10827416,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9459326559954206,
+      "bytes": 2314805,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 272700, "loss": 0.6070579923689365, "lr": 0.0003, "grad_norm": 0.14527561808493308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39914524766, "tokens_per_second": 149764.96514063634, "data_wait_s": 1.1429001460783184e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272800, "loss": 0.6262881923466921, "lr": 0.0003, "grad_norm": 0.16025266180820172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39929169684, "tokens_per_second": 148093.05563199896, "data_wait_s": 1.3905999367125332e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272900, "loss": 0.5818833615630865, "lr": 0.0003, "grad_norm": 0.1676216423082919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39943781282, "tokens_per_second": 149702.61720655984, "data_wait_s": 1.1689997336361557e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 273000, "loss": 0.5703057590872049, "lr": 0.0003, "grad_norm": 0.14665260047803771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39958325289, "tokens_per_second": 148488.69058170656, "data_wait_s": 1.4484001439996064e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:665ce75c187ed9155213b0b284c3e120e67ca41fdc77a32a8326bd845034a288
-size 763890
+oid sha256:806cf9017ba603eaa1b80bc7ee0bba9da374b2d5a9a83b7f8770ebd973f0dc79
+size 766008