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": 77431.31712710959,
-      "required_pages_per_second_upper_estimate": 0.005356878184262239
+      "required_token_positions_per_second": 77275.3117808749,
+      "required_pages_per_second_upper_estimate": 0.005346085372427448
     },
     {
       "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": 9678.914640888699,
-      "required_pages_per_second_upper_estimate": 0.004052359441418711
+      "required_token_positions_per_second": 9659.413972609362,
+      "required_pages_per_second_upper_estimate": 0.004044194918830412
     },
     {
       "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": 19357.829281777398,
-      "required_pages_per_second_upper_estimate": 0.004031725339106414
+      "required_token_positions_per_second": 19318.827945218723,
+      "required_pages_per_second_upper_estimate": 0.0040236023892356535
     },
     {
       "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": 29036.743922666094,
-      "required_pages_per_second_upper_estimate": 0.0017407084616230313
+      "required_token_positions_per_second": 28978.241917828083,
+      "required_pages_per_second_upper_estimate": 0.001737201355760382
     },
     {
       "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": 29036.743922666094,
-      "required_pages_per_second_upper_estimate": 0.0022574791208485475
+      "required_token_positions_per_second": 28978.241917828083,
+      "required_pages_per_second_upper_estimate": 0.002252930847295517
     },
     {
       "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": 29036.743922666094,
-      "required_pages_per_second_upper_estimate": 0.37245217381339507
+      "required_token_positions_per_second": 28978.241917828083,
+      "required_pages_per_second_upper_estimate": 0.3717017729098919
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11004405341418533,
-    "producer_p95_seconds": 0.33807739400072023,
-    "gpu_update_seconds": 0.9521728770152482,
-    "producer_batches_per_second": 9.087269770372654,
-    "gpu_batches_per_second": 1.05022945322143,
-    "producer_headroom": 8.652651801469425,
+    "producer_batch_seconds": 0.09374417065612306,
+    "producer_p95_seconds": 0.3318010310031241,
+    "gpu_update_seconds": 0.954095147607637,
+    "producer_batches_per_second": 10.667329957702103,
+    "gpu_batches_per_second": 1.0481134952918143,
+    "producer_headroom": 10.177647750573158,
     "sustainable": true,
-    "buffer_seconds": 60.939064128975886,
+    "buffer_seconds": 60.10799429928113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386402.4407579931,
-    "upload_bytes_per_second": 4846565.170725534,
+    "generation_bytes_per_second": 420664.2883275442,
+    "upload_bytes_per_second": 4838782.288742459,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.91296879820675,
+    "estimated_full_shards_per_hour": 90.26476371164078,
     "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": 1.2242876330019499,
-      "bytes": 7412020,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.1913529729972652,
       "bytes": 5109029,
@@ -562,6 +556,12 @@
       "bytes": 13769456,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 1.0934254949970637,
+      "bytes": 6083579,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 960900, "loss": 0.5122169256210327, "lr": 3e-05, "grad_norm": 0.3369775143763825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140649557561, "tokens_per_second": 153493.75666208475, "data_wait_s": 1.2072996469214559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961000, "loss": 0.5310725569725037, "lr": 3e-05, "grad_norm": 0.35473942127912833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140664331074, "tokens_per_second": 153975.99668299165, "data_wait_s": 1.0764997568912804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961100, "loss": 0.5285568237304688, "lr": 3e-05, "grad_norm": 0.34067894783062574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140679106352, "tokens_per_second": 153709.10084384642, "data_wait_s": 1.057800545822829e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 961200, "loss": 0.5192878246307373, "lr": 3e-05, "grad_norm": 0.35351720056532393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140693943165, "tokens_per_second": 154038.12617002902, "data_wait_s": 1.1238997103646398e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a66a69532d97c4a8099115fd8973c50d3399142638b16f07278fc395015d0cce
-size 770244
+oid sha256:9463d49070cfbbb24ad41b0c0c2db99e623d26270722b05c878a884267a09344
+size 772362