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": 73953.10105672458,
-      "required_pages_per_second_upper_estimate": 0.006410067367175849
+      "required_token_positions_per_second": 75805.76634710687,
+      "required_pages_per_second_upper_estimate": 0.006570651699008935
     },
     {
       "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": 9244.137632090573,
-      "required_pages_per_second_upper_estimate": 0.0036876247136151956
+      "required_token_positions_per_second": 9475.720793388358,
+      "required_pages_per_second_upper_estimate": 0.0037800066991336995
     },
     {
       "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": 18488.275264181146,
-      "required_pages_per_second_upper_estimate": 0.004276937099836852
+      "required_token_positions_per_second": 18951.441586776717,
+      "required_pages_per_second_upper_estimate": 0.004384082477120466
     },
     {
       "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": 27732.41289627172,
-      "required_pages_per_second_upper_estimate": 0.0016887584446573696
+      "required_token_positions_per_second": 28427.16238016508,
+      "required_pages_per_second_upper_estimate": 0.0017310650431576389
     },
     {
       "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": 27732.41289627172,
-      "required_pages_per_second_upper_estimate": 0.002134917912987279
+      "required_token_positions_per_second": 28427.16238016508,
+      "required_pages_per_second_upper_estimate": 0.0021884016514470512
     },
     {
       "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": 27732.41289627172,
-      "required_pages_per_second_upper_estimate": 0.34434375375630727
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28427.16238016508,
+      "required_pages_per_second_upper_estimate": 0.3529702171693145
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.38717848794527754,
-    "producer_p95_seconds": 0.34330169900204055,
-    "gpu_update_seconds": 0.9969561647380828,
-    "producer_batches_per_second": 2.582788122622496,
-    "gpu_batches_per_second": 1.00305312848205,
-    "producer_headroom": 2.5749265410607967,
+    "producer_batch_seconds": 0.4087763556013897,
+    "producer_p95_seconds": 0.3367987589990662,
+    "gpu_update_seconds": 0.9725909195668178,
+    "producer_batches_per_second": 2.446325444946064,
+    "gpu_batches_per_second": 1.0281815096992577,
+    "producer_headroom": 2.3792739140597967,
     "sustainable": true,
-    "buffer_seconds": 60.81432604902305,
+    "buffer_seconds": 60.3006370131427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377041.7818691066,
-    "upload_bytes_per_second": 5828027.316061261,
+    "generation_bytes_per_second": 378014.0289012829,
+    "upload_bytes_per_second": 6055817.387022032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.90438930563829,
+    "estimated_full_shards_per_hour": 81.11301088599076,
     "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": 3.2236978209984954,
-      "bytes": 7889017,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9988030870008515,
       "bytes": 1104863,
@@ -562,6 +556,12 @@
       "bytes": 823542,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.2165515029992093,
+      "bytes": 11904873,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -273,3 +273,4 @@
 {"step": 27300, "loss": 0.6365917362272739, "lr": 0.0003, "grad_norm": 0.13074144504816781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3995103119, "tokens_per_second": 148590.62108850246, "data_wait_s": 1.2453001545509323e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27400, "loss": 0.6314107235521078, "lr": 0.0003, "grad_norm": 0.13512946290810118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4009838097, "tokens_per_second": 150100.6244749579, "data_wait_s": 1.3975997717352584e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27500, "loss": 0.6400321256369352, "lr": 0.0003, "grad_norm": 0.13620576054388384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4024296085, "tokens_per_second": 146132.9076972112, "data_wait_s": 1.1100000847363845e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 27600, "loss": 0.6811262108385563, "lr": 0.0003, "grad_norm": 0.15013928061374024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4038870735, "tokens_per_second": 149010.26431666, "data_wait_s": 1.4606001059291884e-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:f1e4929772c2fcd18cd5179e73002b9046fbf7356a200774d7ff3faba15db247
-size 580756
+oid sha256:a2b86fb63e0c6ce4b8ba1f876e22c50a8e485c343075e41137385caaeff4269d
+size 582874