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": 72907.62117897434,
-      "required_pages_per_second_upper_estimate": 0.006319447821116983
+      "required_token_positions_per_second": 74879.8959740653,
+      "required_pages_per_second_upper_estimate": 0.0064903996016706985
     },
     {
       "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": 9113.452647371792,
-      "required_pages_per_second_upper_estimate": 0.003635492519296231
+      "required_token_positions_per_second": 9359.986996758162,
+      "required_pages_per_second_upper_estimate": 0.0037338387572834536
     },
     {
       "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": 18226.905294743585,
-      "required_pages_per_second_upper_estimate": 0.004216473757361826
+      "required_token_positions_per_second": 18719.973993516323,
+      "required_pages_per_second_upper_estimate": 0.004330536523110181
     },
     {
       "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": 27340.357942115377,
-      "required_pages_per_second_upper_estimate": 0.001664884354902623
+      "required_token_positions_per_second": 28079.96099027448,
+      "required_pages_per_second_upper_estimate": 0.0017099223001382116
     },
     {
       "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": 27340.357942115377,
-      "required_pages_per_second_upper_estimate": 0.00210473643733875
+      "required_token_positions_per_second": 28079.96099027448,
+      "required_pages_per_second_upper_estimate": 0.0021616731273383116
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27340.357942115377,
-      "required_pages_per_second_upper_estimate": 0.33947574334921066
+      "required_token_positions_per_second": 28079.96099027448,
+      "required_pages_per_second_upper_estimate": 0.3486591379151754
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.22644132698457042,
-    "producer_p95_seconds": 0.42201981900143437,
-    "gpu_update_seconds": 1.0112523054210174,
-    "producer_batches_per_second": 4.41615500720034,
-    "gpu_batches_per_second": 0.9888729001054463,
-    "producer_headroom": 4.465846932127913,
+    "producer_batch_seconds": 0.1397187169919789,
+    "producer_p95_seconds": 0.4319968380004866,
+    "gpu_update_seconds": 0.9846167524796743,
+    "producer_batches_per_second": 7.157237208651214,
+    "gpu_batches_per_second": 1.0156235890579601,
+    "producer_headroom": 7.047135657108846,
     "sustainable": true,
-    "buffer_seconds": 60.675138325261045,
+    "buffer_seconds": 60.06162190126013,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286019.78012500505,
+    "generation_bytes_per_second": 285823.7011279312,
     "upload_bytes_per_second": 5887553.709458662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.37318661511053,
+    "estimated_full_shards_per_hour": 61.331112626823916,
     "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": [

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 154300, "loss": 0.6565862763673067, "lr": 0.0003, "grad_norm": 0.15910598349115007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22581817392, "tokens_per_second": 149324.08239671055, "data_wait_s": 1.1727999662980437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154400, "loss": 0.6272453535348177, "lr": 0.0003, "grad_norm": 0.13534144886402907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22596594148, "tokens_per_second": 149704.79556514946, "data_wait_s": 1.288699422730133e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154500, "loss": 0.5818986315280199, "lr": 0.0003, "grad_norm": 0.15530791020059953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22611260225, "tokens_per_second": 150010.10459833845, "data_wait_s": 1.1598000128287822e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 154600, "loss": 0.6058161295950413, "lr": 0.0003, "grad_norm": 0.14773088963732545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22625939930, "tokens_per_second": 149532.91651421343, "data_wait_s": 1.1928001185879111e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcd9493fa93f09494cc253f522db7e45b20af431bc92bfc621164f9567466095
-size 842256
+oid sha256:076e9278e48be6b6dd0d3957a8198e9a6c4e4933319fb06317beef5deea29352
+size 844374