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": 75657.15877039434,
-      "required_pages_per_second_upper_estimate": 0.0065577707709019535
+      "required_token_positions_per_second": 76061.5145919295,
+      "required_pages_per_second_upper_estimate": 0.006592819308682161
     },
     {
       "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": 9457.144846299292,
-      "required_pages_per_second_upper_estimate": 0.003772596476104712
+      "required_token_positions_per_second": 9507.689323991188,
+      "required_pages_per_second_upper_estimate": 0.003792759423963295
     },
     {
       "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": 18914.289692598584,
-      "required_pages_per_second_upper_estimate": 0.004375488040253361
+      "required_token_positions_per_second": 19015.378647982376,
+      "required_pages_per_second_upper_estimate": 0.00439887319097655
     },
     {
       "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": 28371.43453889788,
-      "required_pages_per_second_upper_estimate": 0.0017276715100058532
+      "required_token_positions_per_second": 28523.067971973567,
+      "required_pages_per_second_upper_estimate": 0.0017369051905210232
     },
     {
       "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": 28371.43453889788,
-      "required_pages_per_second_upper_estimate": 0.0021841115679617643
+      "required_token_positions_per_second": 28523.067971973567,
+      "required_pages_per_second_upper_estimate": 0.0021957847293881385
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28371.43453889788,
-      "required_pages_per_second_upper_estimate": 0.3522782638898628
+      "required_token_positions_per_second": 28523.067971973567,
+      "required_pages_per_second_upper_estimate": 0.35416104364420786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5617753296953651,
-    "producer_p95_seconds": 0.3396099250021507,
-    "gpu_update_seconds": 0.9745013056034924,
-    "producer_batches_per_second": 1.7800710482289632,
-    "gpu_batches_per_second": 1.0261658904404616,
-    "producer_headroom": 1.7346815605661017,
+    "producer_batch_seconds": 0.14316595702342738,
+    "producer_p95_seconds": 0.3533188400033396,
+    "gpu_update_seconds": 0.9693206925414406,
+    "producer_batches_per_second": 6.98490074589703,
+    "gpu_batches_per_second": 1.0316503172733495,
+    "producer_headroom": 6.770608828346134,
     "sustainable": true,
-    "buffer_seconds": 60.419080947416525,
+    "buffer_seconds": 60.09788293756932,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377985.90513097576,
-    "upload_bytes_per_second": 4606749.814687986,
+    "generation_bytes_per_second": 379220.7660240618,
+    "upload_bytes_per_second": 4628569.859332334,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.10697617957072,
+    "estimated_full_shards_per_hour": 81.37194858113662,
     "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": 1.1507087749996572,
-      "bytes": 10108802,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.8633549010010029,
       "bytes": 2675452,
@@ -562,6 +556,12 @@
       "bytes": 9379394,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.4798621720037772,
+      "bytes": 13649971,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 188300, "loss": 0.6326652783900499, "lr": 0.0003, "grad_norm": 0.19723724846528282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27559049208, "tokens_per_second": 151113.9929722391, "data_wait_s": 1.4320998161565512e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 188400, "loss": 0.6406442746520042, "lr": 0.0003, "grad_norm": 0.1704123749444385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27573697255, "tokens_per_second": 150132.10483187786, "data_wait_s": 8.551993232686073e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 188500, "loss": 0.6497993152588606, "lr": 0.0003, "grad_norm": 0.16283055332802274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27588149914, "tokens_per_second": 147988.9902466689, "data_wait_s": 1.0895993909798563e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 188600, "loss": 0.6287831570953131, "lr": 0.0003, "grad_norm": 0.14656467815856106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27602632371, "tokens_per_second": 148410.82226704736, "data_wait_s": 7.88000033935532e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d0fc043c80844feb5b88805d7ec51cb9da8afbfd050e43721ecd620b833ea9c
-size 714393
+oid sha256:e79bc15c7a8ebe1d8aba36b87af67fb22af4b3678c1347d3d7e4c457aa8351f5
+size 716511