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": 76436.01882807349,
-      "required_pages_per_second_upper_estimate": 0.005288021138524693
+      "required_token_positions_per_second": 76112.66634205403,
+      "required_pages_per_second_upper_estimate": 0.0052656508632607305
     },
     {
       "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": 9554.502353509186,
-      "required_pages_per_second_upper_estimate": 0.0040002706147169
+      "required_token_positions_per_second": 9514.083292756754,
+      "required_pages_per_second_upper_estimate": 0.003983347998025825
     },
     {
       "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": 19109.004707018372,
-      "required_pages_per_second_upper_estimate": 0.003979901742129417
+      "required_token_positions_per_second": 19028.166585513507,
+      "required_pages_per_second_upper_estimate": 0.003963065293264578
     },
     {
       "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": 28663.507060527558,
-      "required_pages_per_second_upper_estimate": 0.0017183334816375164
+      "required_token_positions_per_second": 28542.249878270257,
+      "required_pages_per_second_upper_estimate": 0.001711064298710188
     },
     {
       "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": 28663.507060527558,
-      "required_pages_per_second_upper_estimate": 0.0022284615965127335
+      "required_token_positions_per_second": 28542.249878270257,
+      "required_pages_per_second_upper_estimate": 0.0022190343839462064
     },
     {
       "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": 28663.507060527558,
-      "required_pages_per_second_upper_estimate": 0.3676646920964015
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28542.249878270257,
+      "required_pages_per_second_upper_estimate": 0.3661093351582234
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.048346106789075804,
-    "producer_p95_seconds": 0.16380234000007476,
-    "gpu_update_seconds": 0.9645714302027608,
-    "producer_batches_per_second": 20.684188788205756,
-    "gpu_batches_per_second": 1.036729856066535,
-    "producer_headroom": 19.951377562023538,
+    "producer_batch_seconds": 0.371828378257824,
+    "producer_p95_seconds": 0.17321131200060336,
+    "gpu_update_seconds": 0.9686692576063856,
+    "producer_batches_per_second": 2.689412800296283,
+    "gpu_batches_per_second": 1.0323441072869741,
+    "producer_headroom": 2.605151500660111,
     "sustainable": true,
-    "buffer_seconds": 60.76800010277393,
+    "buffer_seconds": 60.0574939715959,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419038.5402616941,
-    "upload_bytes_per_second": 3660309.7080231323,
+    "generation_bytes_per_second": 386532.67968566227,
+    "upload_bytes_per_second": 3914118.0376170482,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.9159160221874,
+    "estimated_full_shards_per_hour": 82.94091504027749,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 7616382,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.8874110320002728,
+      "bytes": 11916705,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -52,3 +52,4 @@
 {"step": 805200, "loss": 0.5211919546127319, "lr": 3e-05, "grad_norm": 0.2688684128774359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117857862950, "tokens_per_second": 149835.87029157692, "data_wait_s": 1.3558999853557907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805300, "loss": 0.5562940239906311, "lr": 3e-05, "grad_norm": 0.29062744684689046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117872513238, "tokens_per_second": 151925.33232381212, "data_wait_s": 1.9110999346594326e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805400, "loss": 0.5579742193222046, "lr": 3e-05, "grad_norm": 0.28295588344758865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117887288118, "tokens_per_second": 145349.65456967396, "data_wait_s": 1.2562999472720549e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 805500, "loss": 0.5162538886070251, "lr": 3e-05, "grad_norm": 0.2728224342692626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117902043224, "tokens_per_second": 152878.94299546952, "data_wait_s": 8.201000127883162e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39914e6214d4b03df477210cc08a1323203878841b6ece76abf830061e6f0dd3
-size 115320
+oid sha256:ea77eb89ce3637d7df0f5e3c998b3bde89fce20ad375c4bc65c5632ef922b1d8
+size 117438