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": 76366.66539233878,
-      "required_pages_per_second_upper_estimate": 0.005283223106918424
+      "required_token_positions_per_second": 75595.71354751004,
+      "required_pages_per_second_upper_estimate": 0.005229886869438436
     },
     {
       "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": 9545.833174042347,
-      "required_pages_per_second_upper_estimate": 0.003996641010307188
+      "required_token_positions_per_second": 9449.464193438755,
+      "required_pages_per_second_upper_estimate": 0.003956293330541618
     },
     {
       "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": 19091.666348084695,
-      "required_pages_per_second_upper_estimate": 0.003976290619206805
+      "required_token_positions_per_second": 18898.92838687751,
+      "required_pages_per_second_upper_estimate": 0.003936148384729192
     },
     {
       "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": 28637.499522127044,
-      "required_pages_per_second_upper_estimate": 0.0017167743694216173
+      "required_token_positions_per_second": 28348.392580316267,
+      "required_pages_per_second_upper_estimate": 0.0016994428497008013
     },
     {
       "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": 28637.499522127044,
-      "required_pages_per_second_upper_estimate": 0.002226439624797165
+      "required_token_positions_per_second": 28348.392580316267,
+      "required_pages_per_second_upper_estimate": 0.0022039628317184223
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28637.499522127044,
-      "required_pages_per_second_upper_estimate": 0.3673310953185188
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28348.392580316267,
+      "required_pages_per_second_upper_estimate": 0.3636227418878191
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07417355280469451,
-    "producer_p95_seconds": 0.13559720100010964,
-    "gpu_update_seconds": 0.9654474189912252,
-    "producer_batches_per_second": 13.48189431660484,
-    "gpu_batches_per_second": 1.0357891898917477,
-    "producer_headroom": 13.016060071078613,
+    "producer_batch_seconds": 0.8029514016406409,
+    "producer_p95_seconds": 0.1484304590001102,
+    "gpu_update_seconds": 0.9752933935025796,
+    "producer_batches_per_second": 1.2454053856269969,
+    "gpu_batches_per_second": 1.0253324862672262,
+    "producer_headroom": 1.2146356448345426,
     "sustainable": true,
-    "buffer_seconds": 60.82318739644719,
+    "buffer_seconds": 60.46819039715994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 72,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305115.52540795895,
-    "upload_bytes_per_second": 4276828.814235062,
+    "generation_bytes_per_second": 257328.00037895291,
+    "upload_bytes_per_second": 4298248.895481772,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.47068902663304,
+    "estimated_full_shards_per_hour": 55.21659859205666,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 7764027,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.4264037609998468,
+      "bytes": 6430937,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 605500, "loss": 0.6408671140670776, "lr": 0.0001, "grad_norm": 0.15215204597776946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88626992392, "tokens_per_second": 150176.7151277495, "data_wait_s": 1.2812000022677239e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605600, "loss": 0.5582746863365173, "lr": 0.0001, "grad_norm": 0.15145503709207161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88641492582, "tokens_per_second": 150708.883809944, "data_wait_s": 1.5875999906711513e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605700, "loss": 0.5830405354499817, "lr": 0.0001, "grad_norm": 0.13374721570454232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88656130069, "tokens_per_second": 152483.22327389175, "data_wait_s": 1.1977999974988052e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 605800, "loss": 0.5844405889511108, "lr": 0.0001, "grad_norm": 0.14258260715490792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88670798435, "tokens_per_second": 120628.73178277216, "data_wait_s": 1.3781999996353989e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72973c9e27fa1ddeade4ccb533d8fed80af7aa288b4aa010f0a605e05307f986
-size 28482
+oid sha256:74ca2336e92558a99c9482bac72f53d07206ac7bf761acc0ed015dd4b60b1db0
+size 30600