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": 77305.13205979759,
-      "required_pages_per_second_upper_estimate": 0.0053481484085158995
+      "required_token_positions_per_second": 77506.21821978375,
+      "required_pages_per_second_upper_estimate": 0.005362060015647917
     },
     {
       "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": 9663.141507474698,
-      "required_pages_per_second_upper_estimate": 0.004045755559838749
+      "required_token_positions_per_second": 9688.27727747297,
+      "required_pages_per_second_upper_estimate": 0.00405627938184246
     },
     {
       "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": 19326.283014949397,
-      "required_pages_per_second_upper_estimate": 0.004025155083657143
+      "required_token_positions_per_second": 19376.55455494594,
+      "required_pages_per_second_upper_estimate": 0.00403562531968876
     },
     {
       "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": 28989.424522424095,
-      "required_pages_per_second_upper_estimate": 0.0017378717358310643
+      "required_token_positions_per_second": 29064.83183241891,
+      "required_pages_per_second_upper_estimate": 0.0017423922889249577
     },
     {
       "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": 28989.424522424095,
-      "required_pages_per_second_upper_estimate": 0.0022538002456157775
+      "required_token_positions_per_second": 29064.83183241891,
+      "required_pages_per_second_upper_estimate": 0.002259662832286171
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28989.424522424095,
-      "required_pages_per_second_upper_estimate": 0.3718452113547042
+      "required_token_positions_per_second": 29064.83183241891,
+      "required_pages_per_second_upper_estimate": 0.3728124553612564
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10220261224981186,
-    "producer_p95_seconds": 0.35881148100452265,
-    "gpu_update_seconds": 0.9537271075737821,
-    "producer_batches_per_second": 9.784485718972812,
-    "gpu_batches_per_second": 1.0485179587103621,
-    "producer_headroom": 9.331729263852917,
+    "producer_batch_seconds": 0.15070655671826216,
+    "producer_p95_seconds": 0.3618435789976502,
+    "gpu_update_seconds": 0.9512527084076029,
+    "producer_batches_per_second": 6.635411370120057,
+    "gpu_batches_per_second": 1.051245364309133,
+    "producer_headroom": 6.311953037225307,
     "sustainable": true,
-    "buffer_seconds": 60.08480777714827,
+    "buffer_seconds": 60.880173338086585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312879.5510812014,
+    "generation_bytes_per_second": 327392.13907791104,
     "upload_bytes_per_second": 4577136.103615089,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.13666819884331,
+    "estimated_full_shards_per_hour": 70.25073174717902,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -438,3 +438,4 @@
 {"step": 688800, "loss": 0.5524390339851379, "lr": 0.0001, "grad_norm": 0.1908495949919119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100821681614, "tokens_per_second": 151533.3032512891, "data_wait_s": 1.2402000720612705e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688900, "loss": 0.5299914479255676, "lr": 0.0001, "grad_norm": 0.18807557853217352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100836258126, "tokens_per_second": 151536.51777184106, "data_wait_s": 1.3773002137895674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689000, "loss": 0.568437933921814, "lr": 0.0001, "grad_norm": 0.20341393785886208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100850908471, "tokens_per_second": 152661.42130221697, "data_wait_s": 1.2404001608956605e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689100, "loss": 0.5370557904243469, "lr": 0.0001, "grad_norm": 0.21108245441357565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100865561638, "tokens_per_second": 151393.064515422, "data_wait_s": 1.2114003766328096e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:073685362598c9dbed2523ea9bb1e013c154c9ccc8e4037ebdd942cf20a27379
-size 938349
+oid sha256:b56910071dc16ab5e5b416e93e7089c8d1e485540c165660a12983b64488066e
+size 940467