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": 75663.82370981443,
-      "required_pages_per_second_upper_estimate": 0.005234598888504004
+      "required_token_positions_per_second": 76336.4807030108,
+      "required_pages_per_second_upper_estimate": 0.005281134860072587
     },
     {
       "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": 9457.977963726804,
-      "required_pages_per_second_upper_estimate": 0.003959857868373483
+      "required_token_positions_per_second": 9542.06008787635,
+      "required_pages_per_second_upper_estimate": 0.003995061297920484
     },
     {
       "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": 18915.955927453608,
-      "required_pages_per_second_upper_estimate": 0.003939694772384751
+      "required_token_positions_per_second": 19084.1201757527,
+      "required_pages_per_second_upper_estimate": 0.003974718950515998
     },
     {
       "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": 28373.933891180408,
-      "required_pages_per_second_upper_estimate": 0.0017009740122877828
+      "required_token_positions_per_second": 28626.180263629056,
+      "required_pages_per_second_upper_estimate": 0.00171609579715817
     },
     {
       "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": 28373.933891180408,
-      "required_pages_per_second_upper_estimate": 0.002205948556293756
+      "required_token_positions_per_second": 28626.180263629056,
+      "required_pages_per_second_upper_estimate": 0.002225559602237103
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28373.933891180408,
-      "required_pages_per_second_upper_estimate": 0.3639503583994614
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28626.180263629056,
+      "required_pages_per_second_upper_estimate": 0.3671859040241795
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04416571465624841,
-    "producer_p95_seconds": 0.136076009999897,
-    "gpu_update_seconds": 0.9744154654774164,
-    "producer_batches_per_second": 22.641997481150767,
-    "gpu_batches_per_second": 1.026256289466884,
-    "producer_headroom": 22.062712514934013,
+    "producer_batch_seconds": 0.06187761550000559,
+    "producer_p95_seconds": 0.13955159800002548,
+    "gpu_update_seconds": 0.9658291726447388,
+    "producer_batches_per_second": 16.160933027548705,
+    "gpu_batches_per_second": 1.03537978384075,
+    "producer_headroom": 15.608700575164399,
     "sustainable": true,
-    "buffer_seconds": 60.41375885959982,
+    "buffer_seconds": 60.84723787661854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 69,
+    "upload_pages_per_commit": 70,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278216.14232363197,
-    "upload_bytes_per_second": 4306545.447321214,
+    "generation_bytes_per_second": 302897.00095115573,
+    "upload_bytes_per_second": 4056491.0351454853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.69870760232658,
+    "estimated_full_shards_per_hour": 64.99464532280926,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 10821331,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.0081327930001862,
+      "bytes": 1129039,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -9,3 +9,4 @@
 {"step": 605300, "loss": 0.5635167956352234, "lr": 0.0001, "grad_norm": 0.13613936133076468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88597999350, "tokens_per_second": 150328.2228645332, "data_wait_s": 7.193999863375211e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 605400, "loss": 0.5422095060348511, "lr": 0.0001, "grad_norm": 0.15087026951664284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88612490448, "tokens_per_second": 150471.31610519273, "data_wait_s": 1.3342000102056772e-05, "gpu_peak_gb": 89.37861251831055}
 {"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}

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 sha256:32f38a7b13a58db997678ccd07139c7b6110dd5e2abd29b7df65e6f7aed8af76
-size 24246
+oid sha256:66c7f98b1a927eca36969e6dbeb1c35b7800c071e57de35c53b7ea48e8fccddb
+size 26364