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": 76839.57816559378,
-      "required_pages_per_second_upper_estimate": 0.005315940309881018
+      "required_token_positions_per_second": 76856.80892439833,
+      "required_pages_per_second_upper_estimate": 0.005317132373755986
     },
     {
       "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": 9604.947270699222,
-      "required_pages_per_second_upper_estimate": 0.00402139084813471
+      "required_token_positions_per_second": 9607.10111554979,
+      "required_pages_per_second_upper_estimate": 0.004022292617996248
     },
     {
       "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": 19209.894541398444,
-      "required_pages_per_second_upper_estimate": 0.004000914433986933
+      "required_token_positions_per_second": 19214.20223109958,
+      "required_pages_per_second_upper_estimate": 0.004001811612150263
     },
     {
       "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": 28814.841812097668,
-      "required_pages_per_second_upper_estimate": 0.0017274057689193585
+      "required_token_positions_per_second": 28821.303346649373,
+      "required_pages_per_second_upper_estimate": 0.0017277931280495413
     },
     {
       "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": 28814.841812097668,
-      "required_pages_per_second_upper_estimate": 0.0022402272077960855
+      "required_token_positions_per_second": 28821.303346649373,
+      "required_pages_per_second_upper_estimate": 0.0022407295636861935
     },
     {
       "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": 28814.841812097668,
-      "required_pages_per_second_upper_estimate": 0.3696058517989465
+      "required_token_positions_per_second": 28821.303346649373,
+      "required_pages_per_second_upper_estimate": 0.3696887334263205
     }
   ],
   "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": 0.42169215313296604,
-    "producer_p95_seconds": 0.2812345700003789,
-    "gpu_update_seconds": 0.9595055277517512,
-    "producer_batches_per_second": 2.3713981694240456,
-    "gpu_batches_per_second": 1.0422034798935789,
-    "producer_headroom": 2.2753696520627558,
+    "producer_batch_seconds": 0.10652082011748121,
+    "producer_p95_seconds": 0.27853430200048024,
+    "gpu_update_seconds": 0.9592904133259548,
+    "producer_batches_per_second": 9.387836095301422,
+    "gpu_batches_per_second": 1.0424371870171214,
+    "producer_headroom": 9.005661168098019,
     "sustainable": true,
-    "buffer_seconds": 60.44884824836033,
+    "buffer_seconds": 60.43529603953515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301234.8800606655,
+    "generation_bytes_per_second": 317562.19182567426,
     "upload_bytes_per_second": 3238230.224910751,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.63799287190412,
+    "estimated_full_shards_per_hour": 68.14145389630957,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 829300, "loss": 0.4824495017528534, "lr": 3e-05, "grad_norm": 0.27106723897553164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121386298888, "tokens_per_second": 150986.97134923682, "data_wait_s": 1.4417997590498999e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829400, "loss": 0.5459651947021484, "lr": 3e-05, "grad_norm": 0.28340184518973255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121401004413, "tokens_per_second": 153089.67534885972, "data_wait_s": 8.663999324198812e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 829500, "loss": 0.5509658455848694, "lr": 3e-05, "grad_norm": 0.2958763041286958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121415610041, "tokens_per_second": 151747.01386022512, "data_wait_s": 1.4760000340174884e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 829600, "loss": 0.5377776026725769, "lr": 3e-05, "grad_norm": 0.2788177400631082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121430194259, "tokens_per_second": 150181.48819472408, "data_wait_s": 1.1932999768760055e-05, "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 sha256:45447a265039cba183eb8172753e1547c755e660260e70e69af291dc2287fc2b
-size 628890
+oid sha256:ebd67d219b12ac5f14064e5cce0f9edf988faaeef378de43f782990063ea8de4
+size 631008