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": 76125.74875792781,
-      "required_pages_per_second_upper_estimate": 0.005266555935146266
+      "required_token_positions_per_second": 76578.92585942372,
+      "required_pages_per_second_upper_estimate": 0.005297907778543499
     },
     {
       "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": 9515.718594740976,
-      "required_pages_per_second_upper_estimate": 0.00398403266481763
+      "required_token_positions_per_second": 9572.365732427965,
+      "required_pages_per_second_upper_estimate": 0.004007749638440423
     },
     {
       "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": 19031.437189481952,
-      "required_pages_per_second_upper_estimate": 0.003963746473819579
+      "required_token_positions_per_second": 19144.73146485593,
+      "required_pages_per_second_upper_estimate": 0.003987342683609017
     },
     {
       "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": 28547.15578422293,
-      "required_pages_per_second_upper_estimate": 0.0017113584002811661
+      "required_token_positions_per_second": 28717.097197283896,
+      "required_pages_per_second_upper_estimate": 0.001721546128508658
     },
     {
       "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": 28547.15578422293,
-      "required_pages_per_second_upper_estimate": 0.002219415796555227
+      "required_token_positions_per_second": 28717.097197283896,
+      "required_pages_per_second_upper_estimate": 0.0022326279939274387
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28547.15578422293,
-      "required_pages_per_second_upper_estimate": 0.3661722628522329
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28717.097197283896,
+      "required_pages_per_second_upper_estimate": 0.368352088830106
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3038120652265093,
-    "producer_p95_seconds": 0.2502027350001299,
-    "gpu_update_seconds": 0.968502789173839,
-    "producer_batches_per_second": 3.2915085161428426,
-    "gpu_batches_per_second": 1.0325215489085262,
-    "producer_headroom": 3.187835178473787,
+    "producer_batch_seconds": 0.1291328086016108,
+    "producer_p95_seconds": 0.22516193699993892,
+    "gpu_update_seconds": 0.9627714044375971,
+    "producer_batches_per_second": 7.743965385939309,
+    "gpu_batches_per_second": 1.038668156730465,
+    "producer_headroom": 7.455668430536928,
     "sustainable": true,
-    "buffer_seconds": 60.04717292877802,
+    "buffer_seconds": 60.65459847956862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496567.24246840423,
+    "generation_bytes_per_second": 506372.47776571236,
     "upload_bytes_per_second": 3442233.9417520748,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.5517707399282,
+    "estimated_full_shards_per_hour": 108.65574598053482,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -152,3 +152,4 @@
 {"step": 857700, "loss": 0.5434330105781555, "lr": 3e-05, "grad_norm": 0.3058737382274146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125543257099, "tokens_per_second": 152213.44510028054, "data_wait_s": 1.3621000107377768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857800, "loss": 0.5173053741455078, "lr": 3e-05, "grad_norm": 0.30637968074020516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125557762639, "tokens_per_second": 149577.61291883624, "data_wait_s": 1.5612000424880534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857900, "loss": 0.48786690831184387, "lr": 3e-05, "grad_norm": 0.30166974147862247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125572644208, "tokens_per_second": 154451.55907358092, "data_wait_s": 1.5167001038207673e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 858000, "loss": 0.49318116903305054, "lr": 3e-05, "grad_norm": 0.31133617668927394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125587339788, "tokens_per_second": 153133.82451254097, "data_wait_s": 1.4123999790172093e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ca473d64ced4e146ca137cc63f7ff84df4a56f12535097a05ead9e80162ee03
-size 328686
+oid sha256:2ba0e95b0bf1d40ef738e6419169f83af08e6653748679f88495f46798ff44c4
+size 330804