Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 73289.33935209109,
-      "required_pages_per_second_upper_estimate": 0.006352534184906878
+      "required_token_positions_per_second": 76226.80839679117,
+      "required_pages_per_second_upper_estimate": 0.006607146556754252
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9161.167419011386,
-      "required_pages_per_second_upper_estimate": 0.003654526655102675
+      "required_token_positions_per_second": 9528.351049598896,
+      "required_pages_per_second_upper_estimate": 0.0038010016952285367
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18322.334838022773,
-      "required_pages_per_second_upper_estimate": 0.004238549702696867
+      "required_token_positions_per_second": 19056.70209919779,
+      "required_pages_per_second_upper_estimate": 0.004408432644147335
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27483.50225703416,
-      "required_pages_per_second_upper_estimate": 0.0016736010926609985
+      "required_token_positions_per_second": 28585.053148796687,
+      "required_pages_per_second_upper_estimate": 0.0017406797625784621
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27483.50225703416,
-      "required_pages_per_second_upper_estimate": 0.002115756082950023
+      "required_token_positions_per_second": 28585.053148796687,
+      "required_pages_per_second_upper_estimate": 0.0022005565198859336
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27483.50225703416,
-      "required_pages_per_second_upper_estimate": 0.3412531166672978
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28585.053148796687,
+      "required_pages_per_second_upper_estimate": 0.3549306920893091
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.34689684568752455,
+    "producer_batch_seconds": 0.32516137095308295,
     "producer_p95_seconds": 0.34261003600113327,
-    "gpu_update_seconds": 1.0059853268126968,
-    "producer_batches_per_second": 2.882701334507877,
-    "gpu_batches_per_second": 0.994050284180923,
-    "producer_headroom": 2.8999552440983036,
+    "gpu_update_seconds": 0.9672187718553837,
+    "producer_batches_per_second": 3.075396062788432,
+    "gpu_batches_per_second": 1.0338922579859913,
+    "producer_headroom": 2.9745808028191094,
     "sustainable": true,
-    "buffer_seconds": 60.35911960876181,
+    "buffer_seconds": 60.934782626889174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400004.6162314558,
-    "upload_bytes_per_second": 3089883.2957121325,
+    "generation_bytes_per_second": 399593.1058812843,
+    "upload_bytes_per_second": 3319725.431627655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.83167901237255,
+    "estimated_full_shards_per_hour": 85.74337847069641,
     "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": 12.285631268001453,
-      "bytes": 12643645,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 2.9626744119996147,
       "bytes": 6098299,
@@ -562,6 +556,12 @@
       "bytes": 5825933,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0486931550003646,
+      "bytes": 6833428,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 303100, "loss": 0.5596464416012168, "lr": 0.0003, "grad_norm": 0.1396069368276492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44366233356, "tokens_per_second": 149082.00414015295, "data_wait_s": 1.4781999198021367e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303200, "loss": 0.7135458812117577, "lr": 0.0003, "grad_norm": 0.22248423776876727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44380786988, "tokens_per_second": 148440.79297462487, "data_wait_s": 1.4043998817214742e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303300, "loss": 0.5642993152141571, "lr": 0.0003, "grad_norm": 0.15055635545494098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44395360099, "tokens_per_second": 149010.12785942582, "data_wait_s": 1.1391999578336254e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 303400, "loss": 0.563585702329874, "lr": 0.0003, "grad_norm": 0.15217523672935845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44409861465, "tokens_per_second": 147336.13125779293, "data_wait_s": 1.5006000467110425e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ee0fc533d6cab4053a5fbcd70900df270aad720e393605184b32be886787c2a
-size 492555
+oid sha256:2a19f1f442486372efa1c68fb1714497ca43317f1f9b21e3763b3dda712d3bde
+size 494673