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": 77115.50135080016,
-      "required_pages_per_second_upper_estimate": 0.005335029316063593
+      "required_token_positions_per_second": 77238.81893604905,
+      "required_pages_per_second_upper_estimate": 0.005343560712747331
     },
     {
       "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": 9639.43766885002,
-      "required_pages_per_second_upper_estimate": 0.004035831257599034
+      "required_token_positions_per_second": 9654.852367006131,
+      "required_pages_per_second_upper_estimate": 0.004042285069821496
     },
     {
       "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": 19278.87533770004,
-      "required_pages_per_second_upper_estimate": 0.004015281314710625
+      "required_token_positions_per_second": 19309.704734012263,
+      "required_pages_per_second_upper_estimate": 0.004021702264936606
     },
     {
       "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": 28918.31300655006,
-      "required_pages_per_second_upper_estimate": 0.0017336087090354135
+      "required_token_positions_per_second": 28964.557101018392,
+      "required_pages_per_second_upper_estimate": 0.0017363809718881462
     },
     {
       "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": 28918.31300655006,
-      "required_pages_per_second_upper_estimate": 0.0022482716380430733
+      "required_token_positions_per_second": 28964.557101018392,
+      "required_pages_per_second_upper_estimate": 0.002251866912981711
     },
     {
       "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": 28918.31300655006,
-      "required_pages_per_second_upper_estimate": 0.370933069182669
+      "required_token_positions_per_second": 28964.557101018392,
+      "required_pages_per_second_upper_estimate": 0.3715262387734687
     }
   ],
   "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": 0.0755676885470109,
-    "producer_p95_seconds": 0.26401633199930075,
-    "gpu_update_seconds": 0.9560723681819776,
-    "producer_batches_per_second": 13.233169086254328,
-    "gpu_batches_per_second": 1.0459459276095942,
-    "producer_headroom": 12.651867306847713,
+    "producer_batch_seconds": 0.09817446206216118,
+    "producer_p95_seconds": 0.2670665100013139,
+    "gpu_update_seconds": 0.9545459267191038,
+    "producer_batches_per_second": 10.185948351485028,
+    "gpu_batches_per_second": 1.0476185294060472,
+    "producer_headroom": 9.722955508681203,
     "sustainable": true,
-    "buffer_seconds": 60.23255919546459,
+    "buffer_seconds": 60.136393383303535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 253535.18262311077,
-    "upload_bytes_per_second": 4599465.360815751,
+    "generation_bytes_per_second": 275909.15623805573,
+    "upload_bytes_per_second": 4630155.953714088,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.40274819393151,
+    "estimated_full_shards_per_hour": 59.20368209224943,
     "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": 1.2096843409999565,
-      "bytes": 2710723,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.2958958110029926,
       "bytes": 5955386,
@@ -562,6 +556,12 @@
       "bytes": 1030943,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.2873325260006823,
+      "bytes": 5911958,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -289,3 +289,4 @@
 {"step": 913900, "loss": 0.5273124575614929, "lr": 3e-05, "grad_norm": 0.3295266797638707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133768519253, "tokens_per_second": 153023.1055769892, "data_wait_s": 1.5828001778572798e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914000, "loss": 0.49982258677482605, "lr": 3e-05, "grad_norm": 0.31729965561477075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133783250680, "tokens_per_second": 153413.05644599695, "data_wait_s": 1.2131997209507972e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 914100, "loss": 0.5752699375152588, "lr": 3e-05, "grad_norm": 0.3338416774421297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133797973145, "tokens_per_second": 153718.12544436217, "data_wait_s": 9.086001227842644e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 914200, "loss": 0.5317642092704773, "lr": 3e-05, "grad_norm": 0.3284383063020626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133812466888, "tokens_per_second": 151018.95310672675, "data_wait_s": 9.747000149218366e-06, "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 sha256:1e1a5c4b8250259e67ef6204500e0fdd89170472752943c020645818aff46780
-size 620418
+oid sha256:9afe1fd576428ad9dd8a106b90b04d5b28c7a4d367233faeea428e983e37f754
+size 622536