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": 75865.72601403558,
-      "required_pages_per_second_upper_estimate": 0.005248566958387912
+      "required_token_positions_per_second": 75018.50012019418,
+      "required_pages_per_second_upper_estimate": 0.005189953905217051
     },
     {
       "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": 9483.215751754447,
-      "required_pages_per_second_upper_estimate": 0.003970424403195714
+      "required_token_positions_per_second": 9377.312515024272,
+      "required_pages_per_second_upper_estimate": 0.003926084929487852
     },
     {
       "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": 18966.431503508895,
-      "required_pages_per_second_upper_estimate": 0.003950207503746612
+      "required_token_positions_per_second": 18754.625030048544,
+      "required_pages_per_second_upper_estimate": 0.0039060938010371496
     },
     {
       "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": 28449.647255263346,
-      "required_pages_per_second_upper_estimate": 0.001705512912856942
+      "required_token_positions_per_second": 28131.937545072815,
+      "required_pages_per_second_upper_estimate": 0.001686466700845658
     },
     {
       "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": 28449.647255263346,
-      "required_pages_per_second_upper_estimate": 0.002211834937323311
+      "required_token_positions_per_second": 28131.937545072815,
+      "required_pages_per_second_upper_estimate": 0.0021871344048133247
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28449.647255263346,
-      "required_pages_per_second_upper_estimate": 0.3649215281392407
+      "required_token_positions_per_second": 28131.937545072815,
+      "required_pages_per_second_upper_estimate": 0.3608462891070255
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07488392445327463,
-    "producer_p95_seconds": 0.27135073699901113,
-    "gpu_update_seconds": 0.9718222427128675,
-    "producer_batches_per_second": 13.354000972851932,
-    "gpu_batches_per_second": 1.0289947647303004,
-    "producer_headroom": 12.97771517462678,
+    "producer_batch_seconds": 0.09151066253917861,
+    "producer_p95_seconds": 0.2575612549990183,
+    "gpu_update_seconds": 0.9827975750231404,
+    "producer_batches_per_second": 10.927688339835463,
+    "gpu_batches_per_second": 1.0175035281059324,
+    "producer_headroom": 10.73970560099894,
     "sustainable": true,
-    "buffer_seconds": 60.25297904819779,
+    "buffer_seconds": 60.9334496514347,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309588.50809866656,
+    "generation_bytes_per_second": 308635.189403251,
     "upload_bytes_per_second": 4938243.489647554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.43048698635099,
+    "estimated_full_shards_per_hour": 66.22592698643825,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -263,3 +263,4 @@
 {"step": 911300, "loss": 0.5353308916091919, "lr": 3e-05, "grad_norm": 0.32864428744185187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133388201077, "tokens_per_second": 152774.58289776597, "data_wait_s": 1.0043000656878576e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911400, "loss": 0.5455825924873352, "lr": 3e-05, "grad_norm": 0.32516339196001465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133403029935, "tokens_per_second": 154391.30851956684, "data_wait_s": 1.600700124981813e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911500, "loss": 0.5518276691436768, "lr": 3e-05, "grad_norm": 0.3302731491156894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133417497103, "tokens_per_second": 150564.36361830018, "data_wait_s": 1.2432999938027933e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 911600, "loss": 0.5138594508171082, "lr": 3e-05, "grad_norm": 0.3166163818687345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133431940195, "tokens_per_second": 150725.50715979826, "data_wait_s": 1.540899756946601e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab8673951a9361f4b77e2ad422270eeb1e44c12ab7f51990554b48bfd603e149
-size 565350
+oid sha256:33f2725420b5a3e1a8e54ad19e502b83a19d986944f0b3ada6ed5e3e4f014482
+size 567468