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": 75057.43269391495,
-      "required_pages_per_second_upper_estimate": 0.006505788034584565
+      "required_token_positions_per_second": 74781.62558873222,
+      "required_pages_per_second_upper_estimate": 0.006481881773734042
     },
     {
       "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": 9382.179086739368,
-      "required_pages_per_second_upper_estimate": 0.0037426915137782706
+      "required_token_positions_per_second": 9347.703198591527,
+      "required_pages_per_second_upper_estimate": 0.003728938566535634
     },
     {
       "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": 18764.358173478737,
-      "required_pages_per_second_upper_estimate": 0.0043408040220095975
+      "required_token_positions_per_second": 18695.406397183055,
+      "required_pages_per_second_upper_estimate": 0.004324853242073408
     },
     {
       "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": 28146.537260218105,
-      "required_pages_per_second_upper_estimate": 0.0017139764456790853
+      "required_token_positions_per_second": 28043.109595774582,
+      "required_pages_per_second_upper_estimate": 0.0017076782435574923
     },
     {
       "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": 28146.537260218105,
-      "required_pages_per_second_upper_estimate": 0.0021667983528934824
+      "required_token_positions_per_second": 28043.109595774582,
+      "required_pages_per_second_upper_estimate": 0.0021588362049785206
     },
     {
       "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": 28146.537260218105,
-      "required_pages_per_second_upper_estimate": 0.34948579237143307
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28043.109595774582,
+      "required_pages_per_second_upper_estimate": 0.3482015669291702
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11241408303112621,
-    "producer_p95_seconds": 0.4368193659975077,
-    "gpu_update_seconds": 0.982287794210383,
-    "producer_batches_per_second": 8.895682578517418,
-    "gpu_batches_per_second": 1.018031584932657,
-    "producer_headroom": 8.738120418047608,
+    "producer_batch_seconds": 0.1490801432422586,
+    "producer_p95_seconds": 0.4074540849978803,
+    "gpu_update_seconds": 0.9859106353942247,
+    "producer_batches_per_second": 6.707801443247726,
+    "gpu_batches_per_second": 1.0142907116527264,
+    "producer_headroom": 6.613292783010664,
     "sustainable": true,
-    "buffer_seconds": 60.90184324104375,
+    "buffer_seconds": 60.14054875904771,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384202.9716437039,
-    "upload_bytes_per_second": 5071481.959077407,
+    "generation_bytes_per_second": 433811.2828466246,
+    "upload_bytes_per_second": 5078253.661915997,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.4410139273008,
+    "estimated_full_shards_per_hour": 93.08580268906644,
     "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": 1.0128854309950839,
-      "bytes": 6059900,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.0482589550010744,
       "bytes": 7056720,
@@ -562,6 +556,12 @@
       "bytes": 6949586,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.203377494995948,
+      "bytes": 7585529,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -419,3 +419,4 @@
 {"step": 41900, "loss": 0.6495089046657085, "lr": 0.0003, "grad_norm": 0.13589079386611236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6131418623, "tokens_per_second": 148359.20912636325, "data_wait_s": 8.381997758988291e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 42000, "loss": 0.6386320497840643, "lr": 0.0003, "grad_norm": 0.13030812197352343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6146009357, "tokens_per_second": 147581.40654735648, "data_wait_s": 1.2259995855856687e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 42100, "loss": 0.6266840100288391, "lr": 0.0003, "grad_norm": 0.1313392719859586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6160514075, "tokens_per_second": 147259.96204561682, "data_wait_s": 1.247199543286115e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 42200, "loss": 0.6159834675490856, "lr": 0.0003, "grad_norm": 0.12582849847270944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6175109717, "tokens_per_second": 147744.1308115539, "data_wait_s": 1.2521995813585818e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c6ebb5dcd7ec52f1eb5e5270fdd2be402b8cd71e7a064ce4e815c1f1b6d0e5f
-size 892333
+oid sha256:e5631df2dcc57b47be491fc43fe6bb18ee8bdd46f0ff4ae53efae4560fbea028
+size 894451