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": 76849.52711689792,
-      "required_pages_per_second_upper_estimate": 0.005316628601416987
+      "required_token_positions_per_second": 77061.85969798175,
+      "required_pages_per_second_upper_estimate": 0.005331318262055833
     },
     {
       "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": 9606.19088961224,
-      "required_pages_per_second_upper_estimate": 0.0040219115254038745
+      "required_token_positions_per_second": 9632.732462247719,
+      "required_pages_per_second_upper_estimate": 0.004033023927615287
     },
     {
       "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": 19212.38177922448,
-      "required_pages_per_second_upper_estimate": 0.004001432460033224
+      "required_token_positions_per_second": 19265.464924495438,
+      "required_pages_per_second_upper_estimate": 0.004012488279296484
     },
     {
       "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": 28818.57266883672,
-      "required_pages_per_second_upper_estimate": 0.0017276294280841765
+      "required_token_positions_per_second": 28898.19738674316,
+      "required_pages_per_second_upper_estimate": 0.0017324028083427628
     },
     {
       "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": 28818.57266883672,
-      "required_pages_per_second_upper_estimate": 0.0022405172654972423
+      "required_token_positions_per_second": 28898.19738674316,
+      "required_pages_per_second_upper_estimate": 0.0022467077370823475
     },
     {
       "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": 28818.57266883672,
-      "required_pages_per_second_upper_estimate": 0.3696537072233132
+      "required_token_positions_per_second": 28898.19738674316,
+      "required_pages_per_second_upper_estimate": 0.3706750476102559
     }
   ],
   "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": 0.4991211297343625,
-    "producer_p95_seconds": 0.1881833600000391,
-    "gpu_update_seconds": 0.9593813100222506,
-    "producer_batches_per_second": 2.003521671246838,
-    "gpu_batches_per_second": 1.0423384211818838,
-    "producer_headroom": 1.9221412456187605,
+    "producer_batch_seconds": 0.20184010238279626,
+    "producer_p95_seconds": 0.17430149899973912,
+    "gpu_update_seconds": 0.9567378764145102,
+    "producer_batches_per_second": 4.9544168289385215,
+    "gpu_batches_per_second": 1.04521836612931,
+    "producer_headroom": 4.740078235790953,
     "sustainable": true,
-    "buffer_seconds": 60.441022531401785,
+    "buffer_seconds": 60.27448621411415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362426.48453255143,
-    "upload_bytes_per_second": 4760678.837315659,
+    "generation_bytes_per_second": 339813.16100285825,
+    "upload_bytes_per_second": 4836988.151520957,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.7682867239228,
+    "estimated_full_shards_per_hour": 72.91599390568076,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 2583323,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3519573450003008,
+      "bytes": 11517949,
+      "pages": 22,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -59,3 +59,4 @@
 {"step": 403000, "loss": 0.6471366286277771, "lr": 0.0003, "grad_norm": 0.15152781109351968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58988575960, "tokens_per_second": 150949.84271917783, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 403100, "loss": 0.5741052031517029, "lr": 0.0003, "grad_norm": 0.18420971794976684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59003197466, "tokens_per_second": 130904.75826209583, "data_wait_s": 1.0632999874360394e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 403200, "loss": 0.6168279647827148, "lr": 0.0003, "grad_norm": 0.17651831990613814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59017911201, "tokens_per_second": 153169.97719119757, "data_wait_s": 6.6350003180559725e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 403300, "loss": 0.5866488814353943, "lr": 0.0003, "grad_norm": 0.17743531496723028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59032566431, "tokens_per_second": 152560.7045594227, "data_wait_s": 8.998999874165747e-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:f860e71c23f8588600bc9fa218c4377d403ef424f0e9d80b2d9824e57ae459a6
-size 130146
+oid sha256:fb3310a0611f31a7bcd34856cd67cdef5ffbaa9fb7092ef232af8f948eba2f1b
+size 132264