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": 76091.51401179984,
-      "required_pages_per_second_upper_estimate": 0.005264187495973577
+      "required_token_positions_per_second": 76980.60544569642,
+      "required_pages_per_second_upper_estimate": 0.005325696904346893
     },
     {
       "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": 9511.43925147498,
-      "required_pages_per_second_upper_estimate": 0.003982240993155007
+      "required_token_positions_per_second": 9622.575680712052,
+      "required_pages_per_second_upper_estimate": 0.00402877149528402
     },
     {
       "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": 19022.87850294996,
-      "required_pages_per_second_upper_estimate": 0.0039619639251227065
+      "required_token_positions_per_second": 19245.151361424105,
+      "required_pages_per_second_upper_estimate": 0.004008257499813409
     },
     {
       "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": 28534.31775442494,
-      "required_pages_per_second_upper_estimate": 0.0017105887799973131
+      "required_token_positions_per_second": 28867.727042136157,
+      "required_pages_per_second_upper_estimate": 0.0017305761577090953
     },
     {
       "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": 28534.31775442494,
-      "required_pages_per_second_upper_estimate": 0.0022184176962069587
+      "required_token_positions_per_second": 28867.727042136157,
+      "required_pages_per_second_upper_estimate": 0.0022443388018140347
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28534.31775442494,
-      "required_pages_per_second_upper_estimate": 0.36600759039038677
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28867.727042136157,
+      "required_pages_per_second_upper_estimate": 0.3702842067461443
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09400757119504988,
-    "producer_p95_seconds": 0.34667230599734467,
-    "gpu_update_seconds": 0.9689385335211845,
-    "producer_batches_per_second": 10.637441083603452,
-    "gpu_batches_per_second": 1.0320572104465038,
-    "producer_headroom": 10.307026563964728,
+    "producer_batch_seconds": 0.12099879210131803,
+    "producer_p95_seconds": 0.3541816189972451,
+    "gpu_update_seconds": 0.9577477284458243,
+    "producer_batches_per_second": 8.264545311846192,
+    "gpu_batches_per_second": 1.044116284799485,
+    "producer_headroom": 7.915349499058277,
     "sustainable": true,
-    "buffer_seconds": 60.07418907831344,
+    "buffer_seconds": 60.338106892086934,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320140.7973062424,
+    "generation_bytes_per_second": 321480.50729893456,
     "upload_bytes_per_second": 5611744.20012044,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.69476260557607,
+    "estimated_full_shards_per_hour": 68.98223318315533,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 430400, "loss": 0.5803173184394836, "lr": 0.0003, "grad_norm": 0.15600094089472244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62998840169, "tokens_per_second": 153888.17733639732, "data_wait_s": 7.443995855282992e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 430500, "loss": 0.8290809988975525, "lr": 0.0003, "grad_norm": 0.28382688817238777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63013588175, "tokens_per_second": 152687.49165665102, "data_wait_s": 1.6183999832719564e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430600, "loss": 0.6422554850578308, "lr": 0.0003, "grad_norm": 0.1645814127129983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63028317997, "tokens_per_second": 152778.01857043232, "data_wait_s": 1.6288002370856702e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 430700, "loss": 0.5648614168167114, "lr": 0.0003, "grad_norm": 0.1504450733490528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63042805057, "tokens_per_second": 149627.51538204597, "data_wait_s": 1.2897995475213975e-05, "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:b9540647d05987998d0054ba53caf7f03164131dc2adf27985ed24f0a248e8ad
-size 715176
+oid sha256:d28cc631d7c2bfe61d151ad9d52e1016c543897e36476d6a230c0ad27bc91fd5
+size 717294