Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75257.78878092673,
-      "required_pages_per_second_upper_estimate": 0.0065231543924088855
+      "required_token_positions_per_second": 75419.27635683188,
+      "required_pages_per_second_upper_estimate": 0.006537151726201824
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9407.22359761584,
-      "required_pages_per_second_upper_estimate": 0.0037526821436157016
+      "required_token_positions_per_second": 9427.409544603985,
+      "required_pages_per_second_upper_estimate": 0.0037607346196760747
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18814.44719523168,
-      "required_pages_per_second_upper_estimate": 0.004352391235655467
+      "required_token_positions_per_second": 18854.81908920797,
+      "required_pages_per_second_upper_estimate": 0.004361730562805814
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28221.670792847523,
-      "required_pages_per_second_upper_estimate": 0.001718551683621036
+      "required_token_positions_per_second": 28282.228633811956,
+      "required_pages_per_second_upper_estimate": 0.0017222393384133334
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28221.670792847523,
-      "required_pages_per_second_upper_estimate": 0.002172582340218225
+      "required_token_positions_per_second": 28282.228633811956,
+      "required_pages_per_second_upper_estimate": 0.0021772442504505
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28221.670792847523,
-      "required_pages_per_second_upper_estimate": 0.3504186993909324
+      "required_token_positions_per_second": 28282.228633811956,
+      "required_pages_per_second_upper_estimate": 0.3511706251016546
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07229905620312849,
-    "producer_p95_seconds": 0.1754832350002289,
-    "gpu_update_seconds": 0.979672684971121,
-    "producer_batches_per_second": 13.831439198741968,
-    "gpu_batches_per_second": 1.0207490882829688,
-    "producer_headroom": 13.550283176846355,
+    "producer_batch_seconds": 0.05029131468747394,
+    "producer_p95_seconds": 0.1713132200002292,
+    "gpu_update_seconds": 0.9775750121384097,
+    "producer_batches_per_second": 19.88414910634798,
+    "gpu_batches_per_second": 1.02293940371137,
+    "producer_headroom": 19.438247304000075,
     "sustainable": true,
-    "buffer_seconds": 60.739706468209505,
+    "buffer_seconds": 60.6096507525814,
     "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": 250775.08278026246,
-    "upload_bytes_per_second": 5178962.351219672,
+    "generation_bytes_per_second": 272400.64527585835,
+    "upload_bytes_per_second": 5093472.592298259,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.8104950194922,
+    "estimated_full_shards_per_hour": 58.45083731371701,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 13323724,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.8108393100001194,
+      "bytes": 1203088,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 318300, "loss": 0.5772039219737053, "lr": 0.0003, "grad_norm": 0.14058944304742868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46590477492, "tokens_per_second": 137929.83436077007, "data_wait_s": 1.293299965254846e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318400, "loss": 0.6080475449562073, "lr": 0.0003, "grad_norm": 0.1703554844603877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46605145498, "tokens_per_second": 148761.08692721464, "data_wait_s": 1.060699969457346e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 318500, "loss": 0.605508603155613, "lr": 0.0003, "grad_norm": 0.14967705484341146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46619811868, "tokens_per_second": 148610.7748379242, "data_wait_s": 1.3629000022774562e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 318600, "loss": 0.6292090825736523, "lr": 0.0003, "grad_norm": 0.16122556275773078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46634337336, "tokens_per_second": 147293.07660231058, "data_wait_s": 1.529099972685799e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44466ce2fa5d033c6705d6d7f34bc7aa2c095ceb017734c1ee2e51debcd82474
-size 74295
+oid sha256:36260a2378439f4fdcfeaeaee89a0c1cc3eae9f37856a07119418bfcbd715ecd
+size 76413