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": 77349.42371667462,
-      "required_pages_per_second_upper_estimate": 0.005351212608109456
+      "required_token_positions_per_second": 77497.64391630878,
+      "required_pages_per_second_upper_estimate": 0.005361466825438393
     },
     {
       "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": 9668.677964584327,
-      "required_pages_per_second_upper_estimate": 0.004048073558816179
+      "required_token_positions_per_second": 9687.205489538597,
+      "required_pages_per_second_upper_estimate": 0.004055830646615815
     },
     {
       "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": 19337.355929168654,
-      "required_pages_per_second_upper_estimate": 0.004027461279676629
+      "required_token_positions_per_second": 19374.410979077194,
+      "required_pages_per_second_upper_estimate": 0.004035178869365197
     },
     {
       "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": 29006.033893752985,
-      "required_pages_per_second_upper_estimate": 0.0017388674422811896
+      "required_token_positions_per_second": 29061.61646861579,
+      "required_pages_per_second_upper_estimate": 0.001742199532774537
     },
     {
       "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": 29006.033893752985,
-      "required_pages_per_second_upper_estimate": 0.002255091551179695
+      "required_token_positions_per_second": 29061.61646861579,
+      "required_pages_per_second_upper_estimate": 0.002259412851893369
     },
     {
       "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": 29006.033893752985,
-      "required_pages_per_second_upper_estimate": 0.3720582585363577
+      "required_token_positions_per_second": 29061.61646861579,
+      "required_pages_per_second_upper_estimate": 0.37277121212677866
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 720750,
-    "producer_batch_seconds": 0.1409164557890108,
-    "producer_p95_seconds": 0.33541668399993796,
-    "gpu_update_seconds": 0.953180986455185,
-    "producer_batches_per_second": 7.096403286620155,
-    "gpu_batches_per_second": 1.0491187027543758,
-    "producer_headroom": 6.764156685024417,
+    "producer_batch_seconds": 0.11464850522651204,
+    "producer_p95_seconds": 0.3228709390023141,
+    "gpu_update_seconds": 0.9513579545672423,
+    "producer_batches_per_second": 8.722311712867876,
+    "gpu_batches_per_second": 1.0511290678752818,
+    "producer_headroom": 8.298040630251883,
     "sustainable": true,
-    "buffer_seconds": 60.05040214667665,
+    "buffer_seconds": 60.886909092303505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339410.2594734205,
-    "upload_bytes_per_second": 5615713.596379661,
+    "generation_bytes_per_second": 346888.1172362502,
+    "upload_bytes_per_second": 5657426.298353266,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.829540616531,
+    "estimated_full_shards_per_hour": 74.4341148168147,
     "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": [
-    {
-      "seconds": 1.3678917270008242,
-      "bytes": 1154643,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.2100134370048181,
       "bytes": 12201541,
@@ -562,6 +556,12 @@
       "bytes": 11951303,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.082737166005245,
+      "bytes": 2735426,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -386,3 +386,4 @@
 {"step": 723600, "loss": 0.5499066710472107, "lr": 8e-05, "grad_norm": 0.21619974403789877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105915173698, "tokens_per_second": 151553.16999639873, "data_wait_s": 1.2883996532764286e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723700, "loss": 0.5622493028640747, "lr": 8e-05, "grad_norm": 0.22698854730628956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105929663900, "tokens_per_second": 150495.401400612, "data_wait_s": 1.7907994333654642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723800, "loss": 0.5341401100158691, "lr": 8e-05, "grad_norm": 0.227167872318649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105944292649, "tokens_per_second": 152246.5032538068, "data_wait_s": 2.308800321770832e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723900, "loss": 0.5106531381607056, "lr": 8e-05, "grad_norm": 0.22065503390254373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105958933471, "tokens_per_second": 152128.62594890653, "data_wait_s": 1.9661994883790612e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fad148770f160b9f80be6c39e71a51f693622db4158b161b4f5c424d81f2505e
-size 827430
+oid sha256:3be9cc33aac539c6a6137a4a6a0f1268e15df189f664eb5599d026dcbcffad0b
+size 829548