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": 75597.52551924778,
-      "required_pages_per_second_upper_estimate": 0.005230012225847606
+      "required_token_positions_per_second": 75316.0245631188,
+      "required_pages_per_second_upper_estimate": 0.005210537336530396
     },
     {
       "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": 9449.690689905972,
-      "required_pages_per_second_upper_estimate": 0.003956388159882657
+      "required_token_positions_per_second": 9414.50307038985,
+      "required_pages_per_second_upper_estimate": 0.00394165583839231
     },
     {
       "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": 18899.381379811944,
-      "required_pages_per_second_upper_estimate": 0.003936242731211208
+      "required_token_positions_per_second": 18829.0061407797,
+      "required_pages_per_second_upper_estimate": 0.00392158542484065
     },
     {
       "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": 28349.072069717917,
-      "required_pages_per_second_upper_estimate": 0.0016994835840529099
+      "required_token_positions_per_second": 28243.50921116955,
+      "required_pages_per_second_upper_estimate": 0.0016931552518680872
     },
     {
       "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": 28349.072069717917,
-      "required_pages_per_second_upper_estimate": 0.0022040156590306427
+      "required_token_positions_per_second": 28243.50921116955,
+      "required_pages_per_second_upper_estimate": 0.0021958086111004507
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28349.072069717917,
-      "required_pages_per_second_upper_estimate": 0.36363145764828464
+      "required_token_positions_per_second": 28243.50921116955,
+      "required_pages_per_second_upper_estimate": 0.3622774106433928
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3013323960156207,
-    "producer_p95_seconds": 0.1436520809998001,
-    "gpu_update_seconds": 0.9752700170222929,
-    "producer_batches_per_second": 3.3185943935087594,
-    "gpu_batches_per_second": 1.0253570627068112,
-    "producer_headroom": 3.2365256106473734,
+    "producer_batch_seconds": 0.31428951145309725,
+    "producer_p95_seconds": 0.14616541999976107,
+    "gpu_update_seconds": 0.9789151834243728,
+    "producer_batches_per_second": 3.1817797398855743,
+    "gpu_batches_per_second": 1.021538961630843,
+    "producer_headroom": 3.11469249768604,
     "sustainable": true,
-    "buffer_seconds": 60.46674105538216,
+    "buffer_seconds": 60.69274137231111,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328432.50508567493,
-    "upload_bytes_per_second": 4727436.266310092,
+    "generation_bytes_per_second": 295280.84868495096,
+    "upload_bytes_per_second": 5337662.209202704,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.47397007396398,
+    "estimated_full_shards_per_hour": 63.36039634143254,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 2320780,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0472984960001668,
+      "bytes": 9808086,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 601800, "loss": 0.5576729774475098, "lr": 0.0003, "grad_norm": 0.14129920849856534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88085408367, "tokens_per_second": 150070.21395653248, "data_wait_s": 1.1115999768662732e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601900, "loss": 0.6145690083503723, "lr": 0.0003, "grad_norm": 0.16814787770286502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88100145606, "tokens_per_second": 152749.24062053088, "data_wait_s": 1.8597999769554008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602000, "loss": 0.5783024430274963, "lr": 0.0003, "grad_norm": 0.1514827487704976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88114623848, "tokens_per_second": 147929.7346813375, "data_wait_s": 1.7145000128948595e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 602100, "loss": 0.576166033744812, "lr": 0.0003, "grad_norm": 0.17736680918012135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88129104162, "tokens_per_second": 150394.35689263276, "data_wait_s": 1.9164000150340144e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6221854c294d371bafb257c193864ddc8e35d6dfac2cdb6993afe151a9c5ecc
-size 42525
+oid sha256:1b69fc3c9944a53d0de23a481fb03e7870b59f4896c700b3bf3d5ba9d8bb20ee
+size 44643