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": 76853.57486152544,
-      "required_pages_per_second_upper_estimate": 0.00531690863378238
+      "required_token_positions_per_second": 76530.9245735352,
+      "required_pages_per_second_upper_estimate": 0.005294586938207403
     },
     {
       "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": 9606.69685769068,
-      "required_pages_per_second_upper_estimate": 0.0040221233636725025
+      "required_token_positions_per_second": 9566.3655716919,
+      "required_pages_per_second_upper_estimate": 0.004005237496437836
     },
     {
       "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": 19213.39371538136,
-      "required_pages_per_second_upper_estimate": 0.004001643219648151
+      "required_token_positions_per_second": 19132.7311433838,
+      "required_pages_per_second_upper_estimate": 0.003984843333116132
     },
     {
       "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": 28820.09057307204,
-      "required_pages_per_second_upper_estimate": 0.0017277204241253781
+      "required_token_positions_per_second": 28699.096715075702,
+      "required_pages_per_second_upper_estimate": 0.0017204670270854216
     },
     {
       "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": 28820.09057307204,
-      "required_pages_per_second_upper_estimate": 0.0022406352758749834
+      "required_token_positions_per_second": 28699.096715075702,
+      "required_pages_per_second_upper_estimate": 0.0022312285356115053
     },
     {
       "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": 28820.09057307204,
-      "required_pages_per_second_upper_estimate": 0.3696731772690453
+      "required_token_positions_per_second": 28699.096715075702,
+      "required_pages_per_second_upper_estimate": 0.3681211979717513
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1038090069455393,
-    "producer_p95_seconds": 0.25710514900129056,
-    "gpu_update_seconds": 0.9593307810709248,
-    "producer_batches_per_second": 9.633075485681355,
-    "gpu_batches_per_second": 1.0423933222320616,
-    "producer_headroom": 9.241305829793871,
+    "producer_batch_seconds": 0.5274474924064521,
+    "producer_p95_seconds": 0.2658839270006865,
+    "gpu_update_seconds": 0.9633752683747863,
+    "producer_batches_per_second": 1.8959233182388477,
+    "gpu_batches_per_second": 1.0380170976228191,
+    "producer_headroom": 1.8264856355263652,
     "sustainable": true,
-    "buffer_seconds": 60.437839207468265,
+    "buffer_seconds": 60.69264190761154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300709.59764952475,
-    "upload_bytes_per_second": 4705221.594474397,
+    "generation_bytes_per_second": 267747.74081970355,
+    "upload_bytes_per_second": 4610988.735932948,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.52527949442202,
+    "estimated_full_shards_per_hour": 57.45243233149843,
     "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": 1.2341140790013014,
-      "bytes": 11781395,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 2.1994080930016935,
       "bytes": 6022375,
@@ -562,6 +556,12 @@
       "bytes": 6068293,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0213589359991602,
+      "bytes": 2097340,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 913100, "loss": 0.5355512499809265, "lr": 3e-05, "grad_norm": 0.33200545651987756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133651473372, "tokens_per_second": 152255.45908432166, "data_wait_s": 1.3250999472802505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913200, "loss": 0.5601001381874084, "lr": 3e-05, "grad_norm": 0.3221489317124942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133666160476, "tokens_per_second": 153073.64850700658, "data_wait_s": 1.756399797159247e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 913300, "loss": 0.5021415948867798, "lr": 3e-05, "grad_norm": 0.3269248681995199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133680802070, "tokens_per_second": 123859.29355803684, "data_wait_s": 1.0933003068203107e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 913400, "loss": 0.5328166484832764, "lr": 3e-05, "grad_norm": 0.38551726637670986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133695394689, "tokens_per_second": 152045.93260453452, "data_wait_s": 1.412599885952659e-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:a90a4baa7fc33da1da6602214d436d5764f699ad1e31e706b341da4b6c1133fd
-size 603474
+oid sha256:4e756473f395ca2a59def6a99f71f687876258fd750e94f27d7e39655871fd50
+size 605592