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": 77255.54553202157,
-      "required_pages_per_second_upper_estimate": 0.005344717897467759
+      "required_token_positions_per_second": 76601.26916859495,
+      "required_pages_per_second_upper_estimate": 0.005299453540515585
     },
     {
       "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": 9656.943191502696,
-      "required_pages_per_second_upper_estimate": 0.004043160454376828
+      "required_token_positions_per_second": 9575.15864607437,
+      "required_pages_per_second_upper_estimate": 0.004008918973061503
     },
     {
       "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": 19313.886383005392,
-      "required_pages_per_second_upper_estimate": 0.004022573192144376
+      "required_token_positions_per_second": 19150.31729214874,
+      "required_pages_per_second_upper_estimate": 0.003988506064125937
     },
     {
       "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": 28970.829574508087,
-      "required_pages_per_second_upper_estimate": 0.001736756997096341
+      "required_token_positions_per_second": 28725.47593822311,
+      "required_pages_per_second_upper_estimate": 0.0017220484212343672
     },
     {
       "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": 28970.829574508087,
-      "required_pages_per_second_upper_estimate": 0.00225235457020584
+      "required_token_positions_per_second": 28725.47593822311,
+      "required_pages_per_second_upper_estimate": 0.0022332794041812756
     },
     {
       "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": 28970.829574508087,
-      "required_pages_per_second_upper_estimate": 0.3716066953285372
+      "required_token_positions_per_second": 28725.47593822311,
+      "required_pages_per_second_upper_estimate": 0.36845956232248317
     }
   ],
   "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": 0.08099538612501078,
-    "producer_p95_seconds": 0.1921959009996499,
-    "gpu_update_seconds": 0.9543392580075765,
-    "producer_batches_per_second": 12.346382279807507,
-    "gpu_batches_per_second": 1.047845398383539,
-    "producer_headroom": 11.782637303989388,
+    "producer_batch_seconds": 0.09046511610937102,
+    "producer_p95_seconds": 0.19785067599877948,
+    "gpu_update_seconds": 0.9624905801198795,
+    "producer_batches_per_second": 11.053984596571063,
+    "gpu_batches_per_second": 1.0389712072563335,
+    "producer_headroom": 10.639356046989894,
     "sustainable": true,
-    "buffer_seconds": 60.123373254477315,
+    "buffer_seconds": 60.636906547552414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448795.1176447459,
-    "upload_bytes_per_second": 5337529.840246645,
+    "generation_bytes_per_second": 539194.1073240534,
+    "upload_bytes_per_second": 5396381.570381515,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.30098483092101,
+    "estimated_full_shards_per_hour": 115.6985036353226,
     "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": 1.479639369999859,
-      "bytes": 2607690,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0580149020001954,
       "bytes": 6239919,
@@ -562,6 +556,12 @@
       "bytes": 8291481,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.2685745549988496,
+      "bytes": 6220542,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 894700, "loss": 0.510834276676178, "lr": 3e-05, "grad_norm": 0.3047739946325943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130959092199, "tokens_per_second": 153209.63930389134, "data_wait_s": 1.7791999198379926e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894800, "loss": 0.5474227070808411, "lr": 3e-05, "grad_norm": 0.3286338105441553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130973604934, "tokens_per_second": 151535.40676080232, "data_wait_s": 1.4946999726817012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 894900, "loss": 0.5160253047943115, "lr": 3e-05, "grad_norm": 0.3158318015444769, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130988146495, "tokens_per_second": 151723.9355317458, "data_wait_s": 1.266899926122278e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 895000, "loss": 0.547240674495697, "lr": 3e-05, "grad_norm": 0.4042976704834889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131002760332, "tokens_per_second": 152778.88040975755, "data_wait_s": 1.3955001122667454e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09873df98cf9ff6732bae1e0ce5e4407e4fe5f27233ad9cdb010b842f19e6abe
-size 210630
+oid sha256:a9ac9936eaca37b231518c183e4c6878407bacc85b9c6a315e48b65302be4971
+size 212748