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": 77060.95924074635,
-      "required_pages_per_second_upper_estimate": 0.005331255966334938
+      "required_token_positions_per_second": 77536.06780480912,
+      "required_pages_per_second_upper_estimate": 0.0053641250792005515
     },
     {
       "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": 9632.619905093294,
-      "required_pages_per_second_upper_estimate": 0.00403297680228519
+      "required_token_positions_per_second": 9692.00847560114,
+      "required_pages_per_second_upper_estimate": 0.004057841556582448
     },
     {
       "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": 19265.239810186587,
-      "required_pages_per_second_upper_estimate": 0.004012441393922614
+      "required_token_positions_per_second": 19384.01695120228,
+      "required_pages_per_second_upper_estimate": 0.004037179540032332
     },
     {
       "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": 28897.859715279883,
-      "required_pages_per_second_upper_estimate": 0.0017323825654541338
+      "required_token_positions_per_second": 29076.025426803422,
+      "required_pages_per_second_upper_estimate": 0.0017430633278166803
     },
     {
       "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": 28897.859715279883,
-      "required_pages_per_second_upper_estimate": 0.0022466814846113375
+      "required_token_positions_per_second": 29076.025426803422,
+      "required_pages_per_second_upper_estimate": 0.0022605330850141484
     },
     {
       "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": 28897.859715279883,
-      "required_pages_per_second_upper_estimate": 0.37067071632328835
+      "required_token_positions_per_second": 29076.025426803422,
+      "required_pages_per_second_upper_estimate": 0.37295603477127565
     }
   ],
   "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": 0.05119067837498292,
-    "producer_p95_seconds": 0.16077158200005215,
-    "gpu_update_seconds": 0.9567490558956858,
-    "producer_batches_per_second": 19.534806565265285,
-    "gpu_batches_per_second": 1.0452061528964078,
-    "producer_headroom": 18.689907738422406,
+    "producer_batch_seconds": 0.1344985748358951,
+    "producer_p95_seconds": 0.16534719299988865,
+    "gpu_update_seconds": 0.9508864982114436,
+    "producer_batches_per_second": 7.4350230195384865,
+    "gpu_batches_per_second": 1.0516502252171376,
+    "producer_headroom": 7.069863003170425,
     "sustainable": true,
-    "buffer_seconds": 60.27519052142821,
+    "buffer_seconds": 60.85673588553239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294655.29223382106,
-    "upload_bytes_per_second": 5299760.673765951,
+    "generation_bytes_per_second": 302532.74159638135,
+    "upload_bytes_per_second": 5266333.658349814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.22616648922895,
+    "estimated_full_shards_per_hour": 64.91648374479847,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 5996670,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1166780060002566,
+      "bytes": 4727222,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 400600, "loss": 0.5313172936439514, "lr": 0.0003, "grad_norm": 0.14731284188264282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58636565138, "tokens_per_second": 151862.11140838277, "data_wait_s": 7.24300025467528e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400700, "loss": 0.6032090187072754, "lr": 0.0003, "grad_norm": 0.1455071352752887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58651151433, "tokens_per_second": 151886.13541794656, "data_wait_s": 1.004999967335607e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 400800, "loss": 0.5677890181541443, "lr": 0.0003, "grad_norm": 0.13917895634959468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58665852294, "tokens_per_second": 153968.55408491456, "data_wait_s": 1.1303000064799562e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 400900, "loss": 0.5627261400222778, "lr": 0.0003, "grad_norm": 0.16006537105349405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58680561683, "tokens_per_second": 152460.9203936658, "data_wait_s": 9.41600046644453e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e724ffcc6b6eb9dc124c75d797738c5486306e719a4069d0a316b4bd74bb5023
-size 79314
+oid sha256:7967e4d89fa608a4b5aee1fff4c09e35593447e7d19b82ecbd37a55934ab32a3
+size 81432