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": 75588.08832034944,
-      "required_pages_per_second_upper_estimate": 0.006551783919349517
+      "required_token_positions_per_second": 74674.2780089374,
+      "required_pages_per_second_upper_estimate": 0.006472577157587381
     },
     {
       "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": 9448.51104004368,
-      "required_pages_per_second_upper_estimate": 0.0037691523217024416
+      "required_token_positions_per_second": 9334.284751117175,
+      "required_pages_per_second_upper_estimate": 0.0037235857472144467
     },
     {
       "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": 18897.02208008736,
-      "required_pages_per_second_upper_estimate": 0.00437149348199849
+      "required_token_positions_per_second": 18668.56950223435,
+      "required_pages_per_second_upper_estimate": 0.004318644998740245
     },
     {
       "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": 28345.533120131044,
-      "required_pages_per_second_upper_estimate": 0.0017260942494971947
+      "required_token_positions_per_second": 28002.854253351525,
+      "required_pages_per_second_upper_estimate": 0.001705226904407387
     },
     {
       "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": 28345.533120131044,
-      "required_pages_per_second_upper_estimate": 0.002182117605045377
+      "required_token_positions_per_second": 28002.854253351525,
+      "required_pages_per_second_upper_estimate": 0.0021557372372848736
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28345.533120131044,
-      "required_pages_per_second_upper_estimate": 0.3519566549552509
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28002.854253351525,
+      "required_pages_per_second_upper_estimate": 0.3477017303022403
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10660708839054678,
-    "producer_p95_seconds": 0.1981838659994537,
-    "gpu_update_seconds": 0.9753917798203042,
-    "producer_batches_per_second": 9.380239298315491,
-    "gpu_batches_per_second": 1.0252290625047398,
-    "producer_headroom": 9.149408304324307,
+    "producer_batch_seconds": 0.19504818567186533,
+    "producer_p95_seconds": 0.22588860600080807,
+    "gpu_update_seconds": 0.9873279255699782,
+    "producer_batches_per_second": 5.126938230957586,
+    "gpu_batches_per_second": 1.0128347169180962,
+    "producer_headroom": 5.061969288096767,
     "sustainable": true,
-    "buffer_seconds": 60.474290348858865,
+    "buffer_seconds": 60.22700345976867,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382849.4958701165,
-    "upload_bytes_per_second": 5233959.998237633,
+    "generation_bytes_per_second": 340773.10673214484,
+    "upload_bytes_per_second": 5407201.594732751,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.15058953359242,
+    "estimated_full_shards_per_hour": 73.12197591279276,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 5862459,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2080343199995696,
+      "bytes": 17359027,
+      "pages": 39,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 204000, "loss": 0.6074925512075424, "lr": 0.0003, "grad_norm": 0.14996113502412647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29858060304, "tokens_per_second": 148552.08069573547, "data_wait_s": 1.8466000256012194e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 204100, "loss": 0.5720325587317348, "lr": 0.0003, "grad_norm": 0.14192007998119296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29872737623, "tokens_per_second": 149290.39704512243, "data_wait_s": 1.235000127053354e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 204200, "loss": 0.5973473954945803, "lr": 0.0003, "grad_norm": 0.13532128023898074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29887428513, "tokens_per_second": 137743.20707045824, "data_wait_s": 1.8417000319459476e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 204300, "loss": 0.6236441843211651, "lr": 0.0003, "grad_norm": 0.1589290398850428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29902201561, "tokens_per_second": 149592.67256207936, "data_wait_s": 1.896799949463457e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bd4a60b1073e1f6b0dfe31bb1abe65629da05ea99d4294b9494c9030e1c65b7
-size 195804
+oid sha256:9ca8e1eccc37a57ae2f6a146d83ed1177d595f199f6e73908551bf54f11b2acd
+size 197922