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": 75637.88026067142,
-      "required_pages_per_second_upper_estimate": 0.006556099758540107
+      "required_token_positions_per_second": 73405.81618438152,
+      "required_pages_per_second_upper_estimate": 0.00636263010152198
     },
     {
       "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": 9454.735032583927,
-      "required_pages_per_second_upper_estimate": 0.003771635165383727
+      "required_token_positions_per_second": 9175.72702304769,
+      "required_pages_per_second_upper_estimate": 0.003660334698838236
     },
     {
       "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": 18909.470065167854,
-      "required_pages_per_second_upper_estimate": 0.004374373104269807
+      "required_token_positions_per_second": 18351.45404609538,
+      "required_pages_per_second_upper_estimate": 0.004245285919003906
     },
     {
       "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": 28364.20509775178,
-      "required_pages_per_second_upper_estimate": 0.001727231274970002
+      "required_token_positions_per_second": 27527.18106914307,
+      "required_pages_per_second_upper_estimate": 0.0016762609031534146
     },
     {
       "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": 28364.20509775178,
-      "required_pages_per_second_upper_estimate": 0.0021835550255699635
+      "required_token_positions_per_second": 27527.18106914307,
+      "required_pages_per_second_upper_estimate": 0.0021191185988168573
     },
     {
       "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": 28364.20509775178,
-      "required_pages_per_second_upper_estimate": 0.3521884984262113
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27527.18106914307,
+      "required_pages_per_second_upper_estimate": 0.34179546133011
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12830915231250195,
-    "producer_p95_seconds": 0.14526532800005043,
-    "gpu_update_seconds": 0.9747496855531993,
-    "producer_batches_per_second": 7.793676304278443,
-    "gpu_batches_per_second": 1.0259044089175267,
-    "producer_headroom": 7.596883526898832,
+    "producer_batch_seconds": 0.07677710470309762,
+    "producer_p95_seconds": 0.15278209000007337,
+    "gpu_update_seconds": 1.004389077492296,
+    "producer_batches_per_second": 13.024716207612533,
+    "gpu_batches_per_second": 0.9956301023272234,
+    "producer_headroom": 13.081882696362907,
     "sustainable": true,
-    "buffer_seconds": 60.43448050429836,
+    "buffer_seconds": 60.26334464953776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 71,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 523339.7261321998,
-    "upload_bytes_per_second": 4961499.055574158,
+    "generation_bytes_per_second": 516792.56591462216,
+    "upload_bytes_per_second": 5163041.679531505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.29652250265595,
+    "estimated_full_shards_per_hour": 110.89165433005331,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 6598891,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9556911770000625,
+      "bytes": 7849875,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 83200, "loss": 0.6312891934067011, "lr": 0.0003, "grad_norm": 0.13139326862963113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12175734091, "tokens_per_second": 149885.2611167311, "data_wait_s": 8.77899992701714e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 83300, "loss": 0.7064251564443111, "lr": 0.0003, "grad_norm": 0.16103747510548544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12190469472, "tokens_per_second": 149129.6152356179, "data_wait_s": 1.8023000166067504e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83400, "loss": 0.6056109964847565, "lr": 0.0003, "grad_norm": 0.14335156640890354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12205019252, "tokens_per_second": 147235.80493464944, "data_wait_s": 9.763999969436554e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 83500, "loss": 0.5754732564091682, "lr": 0.0003, "grad_norm": 0.13478105229435486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12219537614, "tokens_per_second": 147043.53269490437, "data_wait_s": 1.0458999895490706e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c258d6ae20d268ad4ef0371d787286bfec26369611f8bbf1f494d590f03544d5
-size 31935
+oid sha256:2569044342301e0a7360abb3de3df0bd75476c75ee098c95154843431d28fc8b
+size 34053