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": 76870.40820216603,
-      "required_pages_per_second_upper_estimate": 0.005318073203346627
+      "required_token_positions_per_second": 76568.25846621703,
+      "required_pages_per_second_upper_estimate": 0.005297169783529698
     },
     {
       "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": 9608.801025270754,
-      "required_pages_per_second_upper_estimate": 0.004023004334698264
+      "required_token_positions_per_second": 9571.032308277128,
+      "required_pages_per_second_upper_estimate": 0.00400719136159353
     },
     {
       "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": 19217.602050541507,
-      "required_pages_per_second_upper_estimate": 0.004002519704880748
+      "required_token_positions_per_second": 19142.064616554257,
+      "required_pages_per_second_upper_estimate": 0.003986787249437298
     },
     {
       "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": 28826.40307581226,
-      "required_pages_per_second_upper_estimate": 0.0017280988490260209
+      "required_token_positions_per_second": 28713.096924831385,
+      "required_pages_per_second_upper_estimate": 0.0017213063182831923
     },
     {
       "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": 28826.40307581226,
-      "required_pages_per_second_upper_estimate": 0.0022411260451972704
+      "required_token_positions_per_second": 28713.096924831385,
+      "required_pages_per_second_upper_estimate": 0.002232316990339595
     },
     {
       "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": 28826.40307581226,
-      "required_pages_per_second_upper_estimate": 0.36975414727635947
+      "required_token_positions_per_second": 28713.096924831385,
+      "required_pages_per_second_upper_estimate": 0.36830077763024316
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0803421605859107,
-    "producer_p95_seconds": 0.1549706519999745,
-    "gpu_update_seconds": 0.9591207035885432,
-    "producer_batches_per_second": 12.446765094532026,
-    "gpu_batches_per_second": 1.0426216390267744,
-    "producer_headroom": 11.937950094868876,
+    "producer_batch_seconds": 0.12099764146875103,
+    "producer_p95_seconds": 0.14790451299995766,
+    "gpu_update_seconds": 0.9629055365354798,
+    "producer_batches_per_second": 8.264623903915194,
+    "gpu_batches_per_second": 1.0385234709502091,
+    "producer_headroom": 7.9580521144634115,
     "sustainable": true,
-    "buffer_seconds": 60.42460432607822,
+    "buffer_seconds": 60.66304880173523,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401616.0301233082,
-    "upload_bytes_per_second": 3811035.8869786924,
+    "generation_bytes_per_second": 408405.790824547,
+    "upload_bytes_per_second": 3737369.016665812,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.17745092176851,
+    "estimated_full_shards_per_hour": 87.6343755107146,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 2043786,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 0.9116111450002791,
+      "bytes": 853572,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -25,3 +25,4 @@
 {"step": 845000, "loss": 0.5380342602729797, "lr": 3e-05, "grad_norm": 0.32369832958366457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123684316210, "tokens_per_second": 151915.52533154108, "data_wait_s": 1.2603999948623823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845100, "loss": 0.5290372371673584, "lr": 3e-05, "grad_norm": 0.30341436897954355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123699004752, "tokens_per_second": 138643.06287528094, "data_wait_s": 1.079499998013489e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845200, "loss": 0.4966509938240051, "lr": 3e-05, "grad_norm": 0.3019810721601624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123713823583, "tokens_per_second": 154940.3224176908, "data_wait_s": 1.1714999800460646e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 845300, "loss": 0.5320813059806824, "lr": 3e-05, "grad_norm": 0.29579668042968593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123728522712, "tokens_per_second": 153232.44099634653, "data_wait_s": 1.904400005514617e-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:eac5411632fc298bee52654fda9b276f0d7f3fcc65af85d6cc08fa1d4d34df59
-size 58134
+oid sha256:2261e5ac60355ea663700a6f14154ce80276fccadc693c4d49cf1a8b43d0d4fe
+size 60252