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": 73286.81074761767,
-      "required_pages_per_second_upper_estimate": 0.006352315011879811
+      "required_token_positions_per_second": 74705.95975957396,
+      "required_pages_per_second_upper_estimate": 0.006475323251435895
     },
     {
       "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": 9160.851343452208,
-      "required_pages_per_second_upper_estimate": 0.0036544005678363686
+      "required_token_positions_per_second": 9338.244969946745,
+      "required_pages_per_second_upper_estimate": 0.0037251655377161103
     },
     {
       "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": 18321.702686904417,
-      "required_pages_per_second_upper_estimate": 0.004238403465661115
+      "required_token_positions_per_second": 18676.48993989349,
+      "required_pages_per_second_upper_estimate": 0.004320477252597745
     },
     {
       "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": 27482.554030356623,
-      "required_pages_per_second_upper_estimate": 0.0016735433506312972
+      "required_token_positions_per_second": 28014.734909840234,
+      "required_pages_per_second_upper_estimate": 0.0017059503740545622
     },
     {
       "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": 27482.554030356623,
-      "required_pages_per_second_upper_estimate": 0.0021156830858355268
+      "required_token_positions_per_second": 28014.734909840234,
+      "required_pages_per_second_upper_estimate": 0.0021566518431091366
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27482.554030356623,
-      "required_pages_per_second_upper_estimate": 0.3412413428654733
+      "required_token_positions_per_second": 28014.734909840234,
+      "required_pages_per_second_upper_estimate": 0.34784924829383057
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.43211702229669413,
-    "producer_p95_seconds": 0.33367446499687503,
-    "gpu_update_seconds": 1.0060200361822496,
-    "producer_batches_per_second": 2.314187936140581,
-    "gpu_batches_per_second": 0.994015987787783,
-    "producer_headroom": 2.328119431248673,
+    "producer_batch_seconds": 0.09741476364857249,
+    "producer_p95_seconds": 0.3116424680010823,
+    "gpu_update_seconds": 0.9869092136327366,
+    "producer_batches_per_second": 10.265384450426206,
+    "gpu_batches_per_second": 1.0132644281626242,
+    "producer_headroom": 10.131002495607849,
     "sustainable": true,
-    "buffer_seconds": 60.36120217093497,
+    "buffer_seconds": 60.20146203159693,
     "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": 266264.8417269639,
-    "upload_bytes_per_second": 3764138.0954640657,
+    "generation_bytes_per_second": 264142.1729807293,
+    "upload_bytes_per_second": 3785771.537944758,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.13423670632064,
+    "estimated_full_shards_per_hour": 56.678761406578154,
     "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": 1.0164730639990012,
-      "bytes": 6038957,
-      "pages": 35,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9485057740021148,
-      "bytes": 880447,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0207911869983946,
       "bytes": 13578331,
@@ -562,6 +550,18 @@
       "bytes": 6983067,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2635136329990928,
+      "bytes": 9856957,
+      "pages": 29,
+      "shards": 3
+    },
+    {
+      "seconds": 0.961533926998527,
+      "bytes": 492258,
+      "pages": 16,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 69800, "loss": 0.6174406055361032, "lr": 0.0003, "grad_norm": 0.14424695000936255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10215705395, "tokens_per_second": 143158.04213653147, "data_wait_s": 8.422000973951072e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 69900, "loss": 0.6463001444935799, "lr": 0.0003, "grad_norm": 0.130460975222244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10230311326, "tokens_per_second": 130485.51989099575, "data_wait_s": 1.857100141933188e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 70000, "loss": 0.6447709798812866, "lr": 0.0003, "grad_norm": 0.13342261601705638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10244777274, "tokens_per_second": 147803.9612738101, "data_wait_s": 6.407000910257921e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 70100, "loss": 0.6081743184477091, "lr": 0.0003, "grad_norm": 0.13843627369027564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10259436304, "tokens_per_second": 139876.96378341693, "data_wait_s": 1.1140997230540961e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b0b64aae1016c59f0bbb9abe8c4474d12af51cfeea63b6b681ac81494d96cb6
-size 640263
+oid sha256:5da7fac51edc24860d8bce2e1182e930af6cccd2c2326c87b901c0d8a8c82585
+size 642381