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": 76497.73995457668,
-      "required_pages_per_second_upper_estimate": 0.005292291149268928
+      "required_token_positions_per_second": 77593.19891818715,
+      "required_pages_per_second_upper_estimate": 0.00536807754218649
     },
     {
       "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": 9562.217494322085,
-      "required_pages_per_second_upper_estimate": 0.004003500783064801
+      "required_token_positions_per_second": 9699.149864773393,
+      "required_pages_per_second_upper_estimate": 0.004060831507099707
     },
     {
       "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": 19124.43498864417,
-      "required_pages_per_second_upper_estimate": 0.00398311546286818
+      "required_token_positions_per_second": 19398.299729546787,
+      "required_pages_per_second_upper_estimate": 0.004040154266099299
     },
     {
       "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": 28686.652482966256,
-      "required_pages_per_second_upper_estimate": 0.0017197210143717026
+      "required_token_positions_per_second": 29097.44959432018,
+      "required_pages_per_second_upper_estimate": 0.0017443476739465002
     },
     {
       "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": 28686.652482966256,
-      "required_pages_per_second_upper_estimate": 0.002230261051301388
+      "required_token_positions_per_second": 29097.44959432018,
+      "required_pages_per_second_upper_estimate": 0.0022621987198036225
     },
     {
       "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": 28686.652482966256,
-      "required_pages_per_second_upper_estimate": 0.36796157672382673
+      "required_token_positions_per_second": 29097.44959432018,
+      "required_pages_per_second_upper_estimate": 0.3732308409887018
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09607130115622908,
-    "producer_p95_seconds": 0.1663938439996855,
-    "gpu_update_seconds": 0.9637931792988745,
-    "producer_batches_per_second": 10.408935737987159,
-    "gpu_batches_per_second": 1.0375670024221013,
-    "producer_headroom": 10.03206126803232,
+    "producer_batch_seconds": 0.0508526298749743,
+    "producer_p95_seconds": 0.1598949059998631,
+    "gpu_update_seconds": 0.9501863697839995,
+    "producer_batches_per_second": 19.664666359607924,
+    "gpu_batches_per_second": 1.0524251155353075,
+    "producer_headroom": 18.685097941249392,
     "sustainable": true,
-    "buffer_seconds": 60.718970295829095,
+    "buffer_seconds": 60.81192766617597,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428610.00574796274,
-    "upload_bytes_per_second": 1735588.9199305926,
+    "generation_bytes_per_second": 428782.8503969729,
+    "upload_bytes_per_second": 1818501.5951148677,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.96972970322763,
+    "estimated_full_shards_per_hour": 92.00681814128771,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 6440094,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.5451833110000734,
+      "bytes": 7247709,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 650000, "loss": 0.5858739614486694, "lr": 0.0001, "grad_norm": 0.18948754480280644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95143376219, "tokens_per_second": 150650.02094595568, "data_wait_s": 1.4997999642218929e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 650100, "loss": 0.5812369585037231, "lr": 0.0001, "grad_norm": 0.2047598707964555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95157911298, "tokens_per_second": 136365.88316368253, "data_wait_s": 9.395000233780593e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 650200, "loss": 0.5408124923706055, "lr": 0.0001, "grad_norm": 0.18690006406968004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95172613887, "tokens_per_second": 153961.18191900602, "data_wait_s": 6.558000677614473e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 650300, "loss": 0.6387020349502563, "lr": 0.0001, "grad_norm": 0.19528463559114845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95187083881, "tokens_per_second": 151118.3570929203, "data_wait_s": 1.4895000276737846e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9d9bed598a9054f666b9518bc5c9700c59c52cc3ccd1e929e41c56b85b19b31
-size 111084
+oid sha256:8752ba467c589c521682e8c86bef248de3191e406d89bfe01bea14f1712245d6
+size 113202