Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76497.43703176158,
-      "required_pages_per_second_upper_estimate": 0.005292270192365702
+      "required_token_positions_per_second": 75991.13328311018,
+      "required_pages_per_second_upper_estimate": 0.0052572429242474465
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9562.179628970198,
-      "required_pages_per_second_upper_estimate": 0.004003484929632683
+      "required_token_positions_per_second": 9498.891660388772,
+      "required_pages_per_second_upper_estimate": 0.003976987578790709
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19124.359257940396,
-      "required_pages_per_second_upper_estimate": 0.003983099690159737
+      "required_token_positions_per_second": 18997.783320777544,
+      "required_pages_per_second_upper_estimate": 0.00395673726048065
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28686.538886910595,
-      "required_pages_per_second_upper_estimate": 0.001719714204461617
+      "required_token_positions_per_second": 28496.674981166318,
+      "required_pages_per_second_upper_estimate": 0.0017083321532176455
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28686.538886910595,
-      "required_pages_per_second_upper_estimate": 0.0022302522197077167
+      "required_token_positions_per_second": 28496.674981166318,
+      "required_pages_per_second_upper_estimate": 0.002215491136159162
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28686.538886910595,
-      "required_pages_per_second_upper_estimate": 0.36796011963559466
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28496.674981166318,
+      "required_pages_per_second_upper_estimate": 0.3655247493126861
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08139059535159276,
-    "producer_p95_seconds": 0.23944010400009574,
-    "gpu_update_seconds": 0.9637969958312235,
-    "producer_batches_per_second": 12.286431812916215,
-    "gpu_batches_per_second": 1.0375628937684678,
-    "producer_headroom": 11.84162607077382,
+    "producer_batch_seconds": 0.09925918256240607,
+    "producer_p95_seconds": 0.21698916500099585,
+    "gpu_update_seconds": 0.9702184559522395,
+    "producer_batches_per_second": 10.074634650263029,
+    "gpu_batches_per_second": 1.0306957096776013,
+    "producer_headroom": 9.774596474661127,
     "sustainable": true,
-    "buffer_seconds": 60.71921073736708,
+    "buffer_seconds": 60.15354426903885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449892.5236825543,
+    "generation_bytes_per_second": 448994.7031055081,
     "upload_bytes_per_second": 4902876.721535928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.53646262032959,
+    "estimated_full_shards_per_hour": 96.34381122468884,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 456600, "loss": 0.572414219379425, "lr": 0.0003, "grad_norm": 0.16415791665607574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66832543961, "tokens_per_second": 152526.55252933496, "data_wait_s": 1.2068998330505565e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456700, "loss": 0.5616434216499329, "lr": 0.0003, "grad_norm": 0.18140111112097893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66847244892, "tokens_per_second": 151990.12268517324, "data_wait_s": 1.2554999557323754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456800, "loss": 0.5900413393974304, "lr": 0.0003, "grad_norm": 0.15120755257622198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66861847308, "tokens_per_second": 151165.7018522815, "data_wait_s": 1.1833999451482669e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 456900, "loss": 0.6693919897079468, "lr": 0.0003, "grad_norm": 0.15900929390157792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66876376233, "tokens_per_second": 150931.19403093614, "data_wait_s": 1.1603002349147573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2650bba4ed19fbda6b5a6141671497ec939f85dd089ff2a148b6e837d5b7f565
-size 354102
+oid sha256:c2b50523ab1e96273b14381ec51466a74384de8bd8e04ad80c6fe0f3fe2e82a1
+size 356220