Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74891.70315754368,
-      "required_pages_per_second_upper_estimate": 0.006491423018409573
+      "required_token_positions_per_second": 75230.43729631366,
+      "required_pages_per_second_upper_estimate": 0.006520783635044328
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9361.46289469296,
-      "required_pages_per_second_upper_estimate": 0.0037344275150362856
+      "required_token_positions_per_second": 9403.804662039207,
+      "required_pages_per_second_upper_estimate": 0.00375131827909654
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18722.92578938592,
-      "required_pages_per_second_upper_estimate": 0.00433121936913478
+      "required_token_positions_per_second": 18807.609324078414,
+      "required_pages_per_second_upper_estimate": 0.004350809414506581
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28084.38868407888,
-      "required_pages_per_second_upper_estimate": 0.0017101919234605856
+      "required_token_positions_per_second": 28211.413986117623,
+      "required_pages_per_second_upper_estimate": 0.001717927097904491
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28084.38868407888,
-      "required_pages_per_second_upper_estimate": 0.002162013983464015
+      "required_token_positions_per_second": 28211.413986117623,
+      "required_pages_per_second_upper_estimate": 0.002171792742843499
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28084.38868407888,
-      "required_pages_per_second_upper_estimate": 0.3487141150536881
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28211.413986117623,
+      "required_pages_per_second_upper_estimate": 0.35029134417867097
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.0883145759453896,
-    "producer_p95_seconds": 0.2765440340008354,
-    "gpu_update_seconds": 0.984461520989906,
-    "producer_batches_per_second": 11.323159164783423,
-    "gpu_batches_per_second": 1.015783734233177,
-    "producer_headroom": 11.147214493773483,
+    "producer_batch_seconds": 0.0718137899687008,
+    "producer_p95_seconds": 0.22699120700053754,
+    "gpu_update_seconds": 0.9800288639770105,
+    "producer_batches_per_second": 13.924902173187604,
+    "gpu_batches_per_second": 1.020378110030296,
+    "producer_headroom": 13.646806057780053,
     "sustainable": true,
-    "buffer_seconds": 60.05215278038427,
+    "buffer_seconds": 60.76178956657465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318938.4005100719,
-    "upload_bytes_per_second": 5747651.376155982,
+    "generation_bytes_per_second": 318908.2480574429,
+    "upload_bytes_per_second": 5788373.674359424,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.43675624348276,
+    "estimated_full_shards_per_hour": 68.43028622906175,
     "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": 1.0351456819989835,
-      "bytes": 2694003,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.6696938850000151,
       "bytes": 11635327,
@@ -562,6 +556,12 @@
       "bytes": 12447049,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.0412157949976972,
+      "bytes": 5733012,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -211,3 +211,4 @@
 {"step": 376100, "loss": 0.6699020639061928, "lr": 0.0003, "grad_norm": 0.16973057572364986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55049925175, "tokens_per_second": 125615.81850441806, "data_wait_s": 1.5583002095809206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376200, "loss": 0.6044797673821449, "lr": 0.0003, "grad_norm": 0.1593471372290677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55064479747, "tokens_per_second": 148061.10062921714, "data_wait_s": 8.700000762473792e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 376300, "loss": 0.636867169290781, "lr": 0.0003, "grad_norm": 0.16422846199396238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55079089404, "tokens_per_second": 148924.64785953984, "data_wait_s": 1.2706997949862853e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 376400, "loss": 0.594700651243329, "lr": 0.0003, "grad_norm": 0.14515521444552787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55093777204, "tokens_per_second": 149790.23010791634, "data_wait_s": 1.3443001080304384e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:939356be143cfb49a210a4a8e29d9439d8bec996d5c1c29cfdb7d17c6ed99323
-size 454431
+oid sha256:467c064a014d47818ce5b8d452484789088ec87f524faaeffdb7b749bae8762c
+size 456549