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": 75733.69571648147,
-      "required_pages_per_second_upper_estimate": 0.005239432796050621
+      "required_token_positions_per_second": 77402.89689040571,
+      "required_pages_per_second_upper_estimate": 0.00535491200634303
     },
     {
       "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": 9466.711964560183,
-      "required_pages_per_second_upper_estimate": 0.003963514612135742
+      "required_token_positions_per_second": 9675.362111300714,
+      "required_pages_per_second_upper_estimate": 0.004050872071465475
     },
     {
       "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": 18933.423929120367,
-      "required_pages_per_second_upper_estimate": 0.003943332896469755
+      "required_token_positions_per_second": 19350.724222601428,
+      "required_pages_per_second_upper_estimate": 0.00403024554265307
     },
     {
       "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": 28400.135893680548,
-      "required_pages_per_second_upper_estimate": 0.001702544782329525
+      "required_token_positions_per_second": 29026.086333902138,
+      "required_pages_per_second_upper_estimate": 0.001740069555449828
     },
     {
       "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": 28400.135893680548,
-      "required_pages_per_second_upper_estimate": 0.002207985646737713
+      "required_token_positions_per_second": 29026.086333902138,
+      "required_pages_per_second_upper_estimate": 0.002256650540199929
     },
     {
       "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": 28400.135893680548,
-      "required_pages_per_second_upper_estimate": 0.3642864495540148
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29026.086333902138,
+      "required_pages_per_second_upper_estimate": 0.3723154697079583
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.437768546015775,
-    "producer_p95_seconds": 0.25799034599913284,
-    "gpu_update_seconds": 0.973516468495212,
-    "producer_batches_per_second": 2.284312130465319,
-    "gpu_batches_per_second": 1.0272039892100893,
-    "producer_headroom": 2.2238154781913715,
+    "producer_batch_seconds": 0.11885987015614319,
+    "producer_p95_seconds": 0.2659491260019422,
+    "gpu_update_seconds": 0.952522488975975,
+    "producer_batches_per_second": 8.413268487390448,
+    "gpu_batches_per_second": 1.0498439790907892,
+    "producer_headroom": 8.013827440032287,
     "sustainable": true,
-    "buffer_seconds": 60.35802104670314,
+    "buffer_seconds": 60.00891680548643,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284256.36270576104,
-    "upload_bytes_per_second": 5651755.369541565,
+    "generation_bytes_per_second": 302741.160941711,
+    "upload_bytes_per_second": 5674135.274033418,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.994798287197334,
+    "estimated_full_shards_per_hour": 64.96120568455217,
     "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": 1.253374073001396,
-      "bytes": 6015964,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0633229350005422,
       "bytes": 4861789,
@@ -562,6 +556,12 @@
       "bytes": 6246650,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.021862665998924,
+      "bytes": 6426718,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -230,3 +230,4 @@
 {"step": 708000, "loss": 0.585803747177124, "lr": 8e-05, "grad_norm": 0.19999829250549198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103631406402, "tokens_per_second": 153242.46500055972, "data_wait_s": 1.530900044599548e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708100, "loss": 0.586965799331665, "lr": 8e-05, "grad_norm": 0.2340161612950941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103645940375, "tokens_per_second": 143789.36842867962, "data_wait_s": 1.3051998394075781e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708200, "loss": 0.5410672426223755, "lr": 8e-05, "grad_norm": 0.21984543150904165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103660477472, "tokens_per_second": 150857.26024239048, "data_wait_s": 1.440499909222126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 708300, "loss": 0.5177313685417175, "lr": 8e-05, "grad_norm": 0.2094146281548363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103675045922, "tokens_per_second": 151721.92447165836, "data_wait_s": 1.831500048865564e-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:d209deb22c6fa7f631bc8c44c8351df512fe5bab88ca282f029e9de23385b728
-size 494673
+oid sha256:c418022c2af949a0cd595c5dadd9571ec366da0c170bfce78332ed1d73f462c4
+size 496791