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": 76323.18074410634,
-      "required_pages_per_second_upper_estimate": 0.00528021473805542
+      "required_token_positions_per_second": 75700.45710958287,
+      "required_pages_per_second_upper_estimate": 0.0052371332721011825
     },
     {
       "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": 9540.397593013293,
-      "required_pages_per_second_upper_estimate": 0.003994365246038162
+      "required_token_positions_per_second": 9462.557138697859,
+      "required_pages_per_second_upper_estimate": 0.003961775073309817
     },
     {
       "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": 19080.795186026586,
-      "required_pages_per_second_upper_estimate": 0.003974026442841924
+      "required_token_positions_per_second": 18925.114277395718,
+      "required_pages_per_second_upper_estimate": 0.003941602215155764
     },
     {
       "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": 28621.19277903988,
-      "required_pages_per_second_upper_estimate": 0.0017157968050724969
+      "required_token_positions_per_second": 28387.67141609358,
+      "required_pages_per_second_upper_estimate": 0.0017017975559303416
     },
     {
       "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": 28621.19277903988,
-      "required_pages_per_second_upper_estimate": 0.002225171847247921
+      "required_token_positions_per_second": 28387.67141609358,
+      "required_pages_per_second_upper_estimate": 0.002207016588430769
     },
     {
       "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": 28621.19277903988,
-      "required_pages_per_second_upper_estimate": 0.36712192992701326
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28387.67141609358,
+      "required_pages_per_second_upper_estimate": 0.3641265686188425
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11069686311719806,
-    "producer_p95_seconds": 0.29024748300071224,
-    "gpu_update_seconds": 0.9659974765358984,
-    "producer_batches_per_second": 9.03367965306542,
-    "gpu_batches_per_second": 1.035199391603005,
-    "producer_headroom": 8.726511748694886,
+    "producer_batch_seconds": 0.1297192818981614,
+    "producer_p95_seconds": 0.2884019189987157,
+    "gpu_update_seconds": 0.9739439207516597,
+    "producer_batches_per_second": 7.708954176797472,
+    "gpu_batches_per_second": 1.02675316175107,
+    "producer_headroom": 7.508089055845013,
     "sustainable": true,
-    "buffer_seconds": 60.8578410217616,
+    "buffer_seconds": 60.384523086602904,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351310.6164203702,
-    "upload_bytes_per_second": 5484961.642034308,
+    "generation_bytes_per_second": 366721.4491703525,
+    "upload_bytes_per_second": 5463945.271407906,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.38308019121484,
+    "estimated_full_shards_per_hour": 78.68988615353518,
     "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": 1.229033948999131,
-      "bytes": 6359372,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9523748860010528,
       "bytes": 6784780,
@@ -562,6 +556,12 @@
       "bytes": 5950136,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.8421170019973943,
+      "bytes": 2576016,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 470000, "loss": 0.591810941696167, "lr": 0.0003, "grad_norm": 0.16283904999133825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68793746250, "tokens_per_second": 151549.12583840796, "data_wait_s": 1.3904002116760239e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 470100, "loss": 0.5753712058067322, "lr": 0.0003, "grad_norm": 0.1778916165960748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68808468826, "tokens_per_second": 138238.32490474574, "data_wait_s": 1.296800110139884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 470200, "loss": 0.5777687430381775, "lr": 0.0003, "grad_norm": 0.16518298862429426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68823201498, "tokens_per_second": 153452.1597451511, "data_wait_s": 1.7800997738959268e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 470300, "loss": 0.593895435333252, "lr": 0.0003, "grad_norm": 0.19112237144961794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68837909629, "tokens_per_second": 152115.31681617472, "data_wait_s": 8.101000275928527e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e4d7cceef40c9771740573d0008177959c72074d83ff424a6b6435fefc403a9
-size 640263
+oid sha256:0b14570382fff1cadca323ce69b01008ec5fac28087f5a1b905dafad4aa684ba
+size 642381