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": 75699.99939782529,
-      "required_pages_per_second_upper_estimate": 0.005237101606539756
+      "required_token_positions_per_second": 77387.57019974591,
+      "required_pages_per_second_upper_estimate": 0.005353851670320363
     },
     {
       "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": 9462.499924728161,
-      "required_pages_per_second_upper_estimate": 0.003961751119015468
+      "required_token_positions_per_second": 9673.446274968239,
+      "required_pages_per_second_upper_estimate": 0.0040500699508002795
     },
     {
       "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": 18924.999849456322,
-      "required_pages_per_second_upper_estimate": 0.003941578382833655
+      "required_token_positions_per_second": 19346.892549936478,
+      "required_pages_per_second_upper_estimate": 0.004029447506284965
     },
     {
       "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": 28387.49977418448,
-      "required_pages_per_second_upper_estimate": 0.0017017872662599732
+      "required_token_positions_per_second": 29020.338824904717,
+      "required_pages_per_second_upper_estimate": 0.0017397250010613707
     },
     {
       "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": 28387.49977418448,
-      "required_pages_per_second_upper_estimate": 0.0022070032440273096
+      "required_token_positions_per_second": 29020.338824904717,
+      "required_pages_per_second_upper_estimate": 0.002256203696656
     },
     {
       "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": 28387.49977418448,
-      "required_pages_per_second_upper_estimate": 0.3641243669807273
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29020.338824904717,
+      "required_pages_per_second_upper_estimate": 0.37224174683373373
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12304506698461637,
-    "producer_p95_seconds": 0.36925218399846926,
-    "gpu_update_seconds": 0.9739498095969346,
-    "producer_batches_per_second": 8.127103544305635,
-    "gpu_batches_per_second": 1.0267469536380383,
-    "producer_headroom": 7.915390949551045,
+    "producer_batch_seconds": 0.15990689707047068,
+    "producer_p95_seconds": 0.370896681000886,
+    "gpu_update_seconds": 0.9527111370689098,
+    "producer_batches_per_second": 6.253638950665786,
+    "gpu_batches_per_second": 1.0496360975442967,
+    "producer_headroom": 5.957911475507224,
     "sustainable": true,
-    "buffer_seconds": 60.38488819500994,
+    "buffer_seconds": 60.020801635341314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304292.7279833421,
+    "generation_bytes_per_second": 298837.00476091955,
     "upload_bytes_per_second": 5392553.106141569,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.29413585305402,
+    "estimated_full_shards_per_hour": 64.12346465225878,
     "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

@@ -426,3 +426,4 @@
 {"step": 482800, "loss": 0.6001877188682556, "lr": 0.0003, "grad_norm": 0.16184486147186183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70665634891, "tokens_per_second": 151296.96019529446, "data_wait_s": 1.2812997738365084e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 482900, "loss": 0.5762990117073059, "lr": 0.0003, "grad_norm": 0.1542558287123174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70680327522, "tokens_per_second": 152236.4964049248, "data_wait_s": 1.4051998732611537e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483000, "loss": 0.5855228304862976, "lr": 0.0003, "grad_norm": 0.14452549252851202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70695102401, "tokens_per_second": 153096.64240842784, "data_wait_s": 1.3914002920500934e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 483100, "loss": 0.5663217306137085, "lr": 0.0003, "grad_norm": 0.17280523039607762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70709519840, "tokens_per_second": 149986.1279342234, "data_wait_s": 1.4061006368137896e-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 sha256:5cca53d4050ffac83140b028c91cde4a60deab005754a6a1154c2493ee013791
-size 912933
+oid sha256:1446b0b8136c27787c9cbd9b97b56372bf85ed35f096f70f5cdac4a79375090c
+size 915051