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": 77114.03206304689,
-      "required_pages_per_second_upper_estimate": 0.005334927667327604
+      "required_token_positions_per_second": 76498.65788931001,
+      "required_pages_per_second_upper_estimate": 0.005292354654123682
     },
     {
       "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": 9639.254007880862,
-      "required_pages_per_second_upper_estimate": 0.004035754362586525
+      "required_token_positions_per_second": 9562.332236163751,
+      "required_pages_per_second_upper_estimate": 0.004003548823077824
     },
     {
       "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": 19278.508015761723,
-      "required_pages_per_second_upper_estimate": 0.004015204811237804
+      "required_token_positions_per_second": 19124.664472327502,
+      "required_pages_per_second_upper_estimate": 0.003983163258267526
     },
     {
       "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": 28917.762023642583,
-      "required_pages_per_second_upper_estimate": 0.0017335756784514125
+      "required_token_positions_per_second": 28686.996708491257,
+      "required_pages_per_second_upper_estimate": 0.0017197416501663245
     },
     {
       "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": 28917.762023642583,
-      "required_pages_per_second_upper_estimate": 0.0022482288015455345
+      "required_token_positions_per_second": 28686.996708491257,
+      "required_pages_per_second_upper_estimate": 0.0022302878133218687
     },
     {
       "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": 28917.762023642583,
-      "required_pages_per_second_upper_estimate": 0.37092600176553125
+      "required_token_positions_per_second": 28686.996708491257,
+      "required_pages_per_second_upper_estimate": 0.3679659920792609
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.6281943969375448,
-    "producer_p95_seconds": 0.33099070000025677,
-    "gpu_update_seconds": 0.9560905846515906,
-    "producer_batches_per_second": 1.5918639275915418,
-    "gpu_batches_per_second": 1.045925999119017,
-    "producer_headroom": 1.5219661132167746,
+    "producer_batch_seconds": 0.10424452923410854,
+    "producer_p95_seconds": 0.3034218669999973,
+    "gpu_update_seconds": 0.963781614400098,
+    "producer_batches_per_second": 9.592829545560484,
+    "gpu_batches_per_second": 1.0375794527087405,
+    "producer_headroom": 9.245392746085242,
     "sustainable": true,
-    "buffer_seconds": 60.233706833050206,
+    "buffer_seconds": 60.718241707206175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243933.70510627492,
-    "upload_bytes_per_second": 3149284.197445145,
+    "generation_bytes_per_second": 232541.66631246396,
+    "upload_bytes_per_second": 3181306.8622468873,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.342494629775864,
+    "estimated_full_shards_per_hour": 49.8980282977146,
     "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": 1.8874185050008236,
-      "bytes": 2469051,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 24.49332847400001,
       "bytes": 9957694,
@@ -562,6 +556,12 @@
       "bytes": 13136869,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.4420257410019985,
+      "bytes": 5442386,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -324,3 +324,4 @@
 {"step": 832400, "loss": 0.5739641785621643, "lr": 3e-05, "grad_norm": 0.284707157600611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121839877898, "tokens_per_second": 116365.84335738016, "data_wait_s": 8.529998012818396e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 832500, "loss": 0.5149006843566895, "lr": 3e-05, "grad_norm": 0.27419829143884333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121854484090, "tokens_per_second": 152229.1883285111, "data_wait_s": 1.3231998309493065e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 832600, "loss": 0.5355899333953857, "lr": 3e-05, "grad_norm": 0.2973308184618766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121869173641, "tokens_per_second": 138589.4877518306, "data_wait_s": 9.17899888008833e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 832700, "loss": 0.561130166053772, "lr": 3e-05, "grad_norm": 0.2903270618468921, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121884005720, "tokens_per_second": 153546.51378122595, "data_wait_s": 2.0450999727472663e-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 sha256:46bc178cec78a4ad65e166259b2cea5efb4102c61f7893f9ce89e17131fdec1f
-size 695331
+oid sha256:bd68329b4e62592d06ec4f44a2c344317b2f9f3e166c3c16bd882ce9b001edc4
+size 697449