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": 77000.16840730303,
-      "required_pages_per_second_upper_estimate": 0.005327050315423163
+      "required_token_positions_per_second": 76905.46569552396,
+      "required_pages_per_second_upper_estimate": 0.005320498562081721
     },
     {
       "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": 9625.02105091288,
-      "required_pages_per_second_upper_estimate": 0.00402979532072197
+      "required_token_positions_per_second": 9613.183211940495,
+      "required_pages_per_second_upper_estimate": 0.004024839064746421
     },
     {
       "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": 19250.04210182576,
-      "required_pages_per_second_upper_estimate": 0.004009276112061575
+      "required_token_positions_per_second": 19226.36642388099,
+      "required_pages_per_second_upper_estimate": 0.004004345092715295
     },
     {
       "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": 28875.063152738636,
-      "required_pages_per_second_upper_estimate": 0.0017310159463381216
+      "required_token_positions_per_second": 28839.549635821484,
+      "required_pages_per_second_upper_estimate": 0.0017288869652249388
     },
     {
       "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": 28875.063152738636,
-      "required_pages_per_second_upper_estimate": 0.002244909152145237
+      "required_token_positions_per_second": 28839.549635821484,
+      "required_pages_per_second_upper_estimate": 0.002242148132412381
     },
     {
       "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": 28875.063152738636,
-      "required_pages_per_second_upper_estimate": 0.370378306496051
+      "required_token_positions_per_second": 28839.549635821484,
+      "required_pages_per_second_upper_estimate": 0.3699227772324814
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.137767387289216,
+    "producer_batch_seconds": 0.11162943876550457,
     "producer_p95_seconds": 0.2926373079972109,
-    "gpu_update_seconds": 0.957504399341123,
-    "producer_batches_per_second": 7.25861192315924,
-    "gpu_batches_per_second": 1.044381624448012,
-    "producer_headroom": 6.950152849534901,
+    "gpu_update_seconds": 0.9586834867094642,
+    "producer_batches_per_second": 8.958210406312794,
+    "gpu_batches_per_second": 1.0430971367122932,
+    "producer_headroom": 8.588088387000957,
     "sustainable": true,
-    "buffer_seconds": 60.322777158490744,
+    "buffer_seconds": 60.39705966269625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323623.7453334474,
+    "generation_bytes_per_second": 323330.3373047826,
     "upload_bytes_per_second": 5758397.724492029,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.4421221733338,
+    "estimated_full_shards_per_hour": 69.37916364057168,
     "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

@@ -318,3 +318,4 @@
 {"step": 428900, "loss": 0.5718140006065369, "lr": 0.0003, "grad_norm": 0.1473230188598794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62779181864, "tokens_per_second": 151555.54729809656, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429000, "loss": 0.5638366341590881, "lr": 0.0003, "grad_norm": 0.15660331808988368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62793871506, "tokens_per_second": 152104.7180156272, "data_wait_s": 8.949999028118327e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 429100, "loss": 0.6631481051445007, "lr": 0.0003, "grad_norm": 0.17248152837054945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62808446446, "tokens_per_second": 150982.356583782, "data_wait_s": 2.0313000277383253e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 429200, "loss": 0.5594878196716309, "lr": 0.0003, "grad_norm": 0.13760300670004028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62822966979, "tokens_per_second": 150400.38467674743, "data_wait_s": 1.567399885971099e-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:e16b12807b1ccc2880371ffd7da8c05fe039d1e8779bdc70472241dd4395209c
-size 682623
+oid sha256:ff3c78a765734295341ff719a21e9cd09463fcc6035390ba1aef2a3673f0942a
+size 684741