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": 76930.64008666214,
-      "required_pages_per_second_upper_estimate": 0.005322240184873295
+      "required_token_positions_per_second": 77068.24010550776,
+      "required_pages_per_second_upper_estimate": 0.005331759673453067
     },
     {
       "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": 9616.330010832768,
-      "required_pages_per_second_upper_estimate": 0.004026156563729982
+      "required_token_positions_per_second": 9633.53001318847,
+      "required_pages_per_second_upper_estimate": 0.004033357845539422
     },
     {
       "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": 19232.660021665535,
-      "required_pages_per_second_upper_estimate": 0.004005655883160481
+      "required_token_positions_per_second": 19267.06002637694,
+      "required_pages_per_second_upper_estimate": 0.004012820496952736
     },
     {
       "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": 28848.9900324983,
-      "required_pages_per_second_upper_estimate": 0.001729452902591063
+      "required_token_positions_per_second": 28900.59003956541,
+      "required_pages_per_second_upper_estimate": 0.0017325462442260874
     },
     {
       "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": 28848.9900324983,
-      "required_pages_per_second_upper_estimate": 0.002242882081730087
+      "required_token_positions_per_second": 28900.59003956541,
+      "required_pages_per_second_upper_estimate": 0.0022468937553151097
     },
     {
       "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": 28848.9900324983,
-      "required_pages_per_second_upper_estimate": 0.3700438685047434
+      "required_token_positions_per_second": 28900.59003956541,
+      "required_pages_per_second_upper_estimate": 0.37070573799162926
     }
   ],
   "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": 0.18851610182812095,
-    "producer_p95_seconds": 0.16934688499986805,
-    "gpu_update_seconds": 0.9583697720043097,
-    "producer_batches_per_second": 5.304586665555748,
-    "gpu_batches_per_second": 1.043438586244875,
-    "producer_headroom": 5.083755513245764,
+    "producer_batch_seconds": 0.0785811692031686,
+    "producer_p95_seconds": 0.16191080499993404,
+    "gpu_update_seconds": 0.9566586689804397,
+    "producer_batches_per_second": 12.725695101513931,
+    "gpu_batches_per_second": 1.0453049059449295,
+    "producer_headroom": 12.17414653766522,
     "sustainable": true,
-    "buffer_seconds": 60.37729563627151,
+    "buffer_seconds": 60.269496145767704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449116.76606956986,
+    "generation_bytes_per_second": 446680.12147636036,
     "upload_bytes_per_second": 1480632.9731717375,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.37000309529611,
+    "estimated_full_shards_per_hour": 95.84715588777645,
     "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

@@ -46,3 +46,4 @@
 {"step": 649600, "loss": 0.5479921698570251, "lr": 0.0001, "grad_norm": 0.19101405004114777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95084832627, "tokens_per_second": 150693.50332473035, "data_wait_s": 9.286000022257213e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 649700, "loss": 0.5493369102478027, "lr": 0.0001, "grad_norm": 0.16725687475794224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95099487377, "tokens_per_second": 152387.97609152904, "data_wait_s": 1.0320999535906594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 649800, "loss": 0.5211204290390015, "lr": 0.0001, "grad_norm": 0.1773628802035087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95114114133, "tokens_per_second": 152121.50754459537, "data_wait_s": 1.2018000234093051e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 649900, "loss": 0.5661001205444336, "lr": 0.0001, "grad_norm": 0.17886634781365532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95128841835, "tokens_per_second": 153152.96961128688, "data_wait_s": 9.450000106880907e-06, "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 sha256:59315ff09c0300cc5317d494eca870537adbc775f2b4f540b962ae9b4f769881
-size 101829
+oid sha256:3bbef15a0505b8695a39d2f8dd858f6502a22a2c4f733bed9899285a496903c7
+size 103947