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": 74879.8959740653,
-      "required_pages_per_second_upper_estimate": 0.0064903996016706985
+      "required_token_positions_per_second": 75618.67924422336,
+      "required_pages_per_second_upper_estimate": 0.00655443546309892
     },
     {
       "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": 9359.986996758162,
-      "required_pages_per_second_upper_estimate": 0.0037338387572834536
+      "required_token_positions_per_second": 9452.33490552792,
+      "required_pages_per_second_upper_estimate": 0.003770677718815988
     },
     {
       "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": 18719.973993516323,
-      "required_pages_per_second_upper_estimate": 0.004330536523110181
+      "required_token_positions_per_second": 18904.66981105584,
+      "required_pages_per_second_upper_estimate": 0.00437326264996258
     },
     {
       "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": 28079.96099027448,
-      "required_pages_per_second_upper_estimate": 0.0017099223001382116
+      "required_token_positions_per_second": 28357.00471658376,
+      "required_pages_per_second_upper_estimate": 0.00172679280953435
     },
     {
       "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": 28079.96099027448,
-      "required_pages_per_second_upper_estimate": 0.0021616731273383116
+      "required_token_positions_per_second": 28357.00471658376,
+      "required_pages_per_second_upper_estimate": 0.0021830007202957185
     },
     {
       "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": 28079.96099027448,
-      "required_pages_per_second_upper_estimate": 0.3486591379151754
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28357.00471658376,
+      "required_pages_per_second_upper_estimate": 0.3520990937902301
     }
   ],
   "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": 688946,
-    "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,
+    "producer_batch_seconds": 0.18729948789814443,
+    "producer_p95_seconds": 0.44816705500124954,
+    "gpu_update_seconds": 0.9749971929803601,
+    "producer_batches_per_second": 5.339042894467556,
+    "gpu_batches_per_second": 1.0256439784644011,
+    "producer_headroom": 5.205551835307604,
     "sustainable": true,
-    "buffer_seconds": 60.06162190126013,
+    "buffer_seconds": 60.44982596478233,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285823.7011279312,
-    "upload_bytes_per_second": 5887553.709458662,
+    "generation_bytes_per_second": 296481.5100437329,
+    "upload_bytes_per_second": 5833093.479117978,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.331112626823916,
+    "estimated_full_shards_per_hour": 63.618030319061184,
     "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": 1.1173031580001407,
-      "bytes": 6247577,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.892657821001194,
       "bytes": 2615407,
@@ -562,6 +556,12 @@
       "bytes": 10339350,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.1235615409968887,
+      "bytes": 2445045,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -394,3 +394,4 @@
 {"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}
+{"step": 154700, "loss": 0.6039471216499805, "lr": 0.0003, "grad_norm": 0.14264851109627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22640513205, "tokens_per_second": 149201.79804365436, "data_wait_s": 1.2743999832309783e-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 sha256:076e9278e48be6b6dd0d3957a8198e9a6c4e4933319fb06317beef5deea29352
-size 844374
+oid sha256:a4b145cdfef9eb21f35f870847e0d5cdf12811a841a970781a7997b978b4b224
+size 846492