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": 76516.94715767716,
-      "required_pages_per_second_upper_estimate": 0.005293619948146271
+      "required_token_positions_per_second": 77541.49989376438,
+      "required_pages_per_second_upper_estimate": 0.005364500883718659
     },
     {
       "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": 9564.618394709645,
-      "required_pages_per_second_upper_estimate": 0.00400450598992057
+      "required_token_positions_per_second": 9692.687486720548,
+      "required_pages_per_second_upper_estimate": 0.004058125844358779
     },
     {
       "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": 19129.23678941929,
-      "required_pages_per_second_upper_estimate": 0.003984115551337635
+      "required_token_positions_per_second": 19385.374973441096,
+      "required_pages_per_second_upper_estimate": 0.004037462380251223
     },
     {
       "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": 28693.855184128937,
-      "required_pages_per_second_upper_estimate": 0.0017201528053085175
+      "required_token_positions_per_second": 29078.062460161644,
+      "required_pages_per_second_upper_estimate": 0.0017431854448561358
     },
     {
       "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": 28693.855184128937,
-      "required_pages_per_second_upper_estimate": 0.002230821029635452
+      "required_token_positions_per_second": 29078.062460161644,
+      "required_pages_per_second_upper_estimate": 0.002260691455397787
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28693.855184128937,
-      "required_pages_per_second_upper_estimate": 0.3680539652406836
+      "required_token_positions_per_second": 29078.062460161644,
+      "required_pages_per_second_upper_estimate": 0.3729821636479989
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05456103344538121,
+    "producer_batch_seconds": 0.07472384027346379,
     "producer_p95_seconds": 0.18589668200002052,
-    "gpu_update_seconds": 0.9635512489549534,
-    "producer_batches_per_second": 18.328098587081538,
-    "gpu_batches_per_second": 1.037827516787071,
-    "producer_headroom": 17.660062284551934,
+    "gpu_update_seconds": 0.9508198848488995,
+    "producer_batches_per_second": 13.382609838310515,
+    "gpu_batches_per_second": 1.0517239026389484,
+    "producer_headroom": 12.724451545440152,
     "sustainable": true,
-    "buffer_seconds": 60.70372868416206,
+    "buffer_seconds": 60.852472630329565,
     "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": 322051.2599236678,
-    "upload_bytes_per_second": 5950780.058231998,
+    "generation_bytes_per_second": 334800.48430817964,
+    "upload_bytes_per_second": 5921654.677950205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.10470340998198,
+    "estimated_full_shards_per_hour": 71.84039017614404,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 11473932,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.2914608989995031,
+      "bytes": 6033641,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 491300, "loss": 0.5596840381622314, "lr": 0.0003, "grad_norm": 0.16264493569802665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71909687015, "tokens_per_second": 153820.41107018662, "data_wait_s": 8.785000318312086e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 491400, "loss": 0.5643754005432129, "lr": 0.0003, "grad_norm": 0.16921761438386776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71924314684, "tokens_per_second": 139882.67433042926, "data_wait_s": 1.0732999726315029e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491500, "loss": 0.664214551448822, "lr": 0.0003, "grad_norm": 0.1895744930643989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71939047513, "tokens_per_second": 154091.79369911706, "data_wait_s": 1.1963998986175284e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 491600, "loss": 0.5676164627075195, "lr": 0.0003, "grad_norm": 0.1384545730428987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71953732155, "tokens_per_second": 153761.22089275956, "data_wait_s": 1.5396000890177675e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7234330e4b1a6f652e887d69eea9d3060cf672cb097a8ba7c665522e71051b8a
-size 138618
+oid sha256:5c2a3d87a223326f71c5a8b0b29e8d4bde1223dbc445c76c001c8113e24668a9
+size 140736