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": 77629.85999693618,
-      "required_pages_per_second_upper_estimate": 0.00537061384067977
+      "required_token_positions_per_second": 77084.59996759445,
+      "required_pages_per_second_upper_estimate": 0.005332891486672334
     },
     {
       "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": 9703.732499617023,
-      "required_pages_per_second_upper_estimate": 0.004062750160612437
+      "required_token_positions_per_second": 9635.574995949306,
+      "required_pages_per_second_upper_estimate": 0.004034214037117288
     },
     {
       "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": 19407.464999234046,
-      "required_pages_per_second_upper_estimate": 0.004042063150070739
+      "required_token_positions_per_second": 19271.149991898612,
+      "required_pages_per_second_upper_estimate": 0.004013672328911256
     },
     {
       "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": 29111.19749885107,
-      "required_pages_per_second_upper_estimate": 0.0017451718398312924
+      "required_token_positions_per_second": 28906.72498784792,
+      "required_pages_per_second_upper_estimate": 0.0017329140250081003
     },
     {
       "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": 29111.19749885107,
-      "required_pages_per_second_upper_estimate": 0.0022632675589102577
+      "required_token_positions_per_second": 28906.72498784792,
+      "required_pages_per_second_upper_estimate": 0.0022473707205592896
     },
     {
       "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": 29111.19749885107,
-      "required_pages_per_second_upper_estimate": 0.37340718434667425
+      "required_token_positions_per_second": 28906.72498784792,
+      "required_pages_per_second_upper_estimate": 0.37078443052100307
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07564898855450508,
-    "producer_p95_seconds": 0.19816767900010746,
-    "gpu_update_seconds": 0.9497376396519306,
-    "producer_batches_per_second": 13.21894739252859,
-    "gpu_batches_per_second": 1.0529223632396945,
-    "producer_headroom": 12.554531895263144,
+    "producer_batch_seconds": 0.15321200974227622,
+    "producer_p95_seconds": 0.20650726899839356,
+    "gpu_update_seconds": 0.9564556348608473,
+    "producer_batches_per_second": 6.526903482841444,
+    "gpu_batches_per_second": 1.0455268007757494,
+    "producer_headroom": 6.242693614356589,
     "sustainable": true,
-    "buffer_seconds": 60.78320893772356,
+    "buffer_seconds": 60.25670499623338,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412275.99405777274,
-    "upload_bytes_per_second": 5413992.046926237,
+    "generation_bytes_per_second": 490167.4260891485,
+    "upload_bytes_per_second": 5428082.149759443,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.46483103084456,
+    "estimated_full_shards_per_hour": 105.17851912504045,
     "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": [
-    {
-      "seconds": 1.545225169999867,
-      "bytes": 7780089,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.7348476629995275,
       "bytes": 1091907,
@@ -562,6 +556,12 @@
       "bytes": 6968332,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.032158955000341,
+      "bytes": 6117925,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 658300, "loss": 0.5375291705131531, "lr": 0.0001, "grad_norm": 0.18677238945245248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96357548252, "tokens_per_second": 153462.8223868825, "data_wait_s": 1.752299976942595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658400, "loss": 0.5696204304695129, "lr": 0.0001, "grad_norm": 0.19955120658093234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96372094579, "tokens_per_second": 152284.3883323311, "data_wait_s": 1.732099917717278e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658500, "loss": 0.5744720697402954, "lr": 0.0001, "grad_norm": 0.17951242339152534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96386594202, "tokens_per_second": 151203.9696837603, "data_wait_s": 1.3013999705435708e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 658600, "loss": 0.5846948027610779, "lr": 0.0001, "grad_norm": 0.1931559684828068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96401125533, "tokens_per_second": 150968.4826024349, "data_wait_s": 1.0240999472443946e-05, "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:c34ae8ff1179da9d8c9ac156e80eff7d028f926fd1283d94e2b7256ad749334d
-size 287661
+oid sha256:71e62698d68ccb1b99b18cf9ebff8eac7e9e9b378695745bb37d304af934bdf5
+size 289779