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": 74832.11508728137,
-      "required_pages_per_second_upper_estimate": 0.006486258075503819
+      "required_token_positions_per_second": 74220.3648076825,
+      "required_pages_per_second_upper_estimate": 0.006433233111735098
     },
     {
       "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": 9354.014385910172,
-      "required_pages_per_second_upper_estimate": 0.0037314561935177007
+      "required_token_positions_per_second": 9277.545600960313,
+      "required_pages_per_second_upper_estimate": 0.003700951651890982
     },
     {
       "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": 18708.028771820344,
-      "required_pages_per_second_upper_estimate": 0.004327773206299538
+      "required_token_positions_per_second": 18555.091201920626,
+      "required_pages_per_second_upper_estimate": 0.0042923937911125394
     },
     {
       "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": 28062.043157730517,
-      "required_pages_per_second_upper_estimate": 0.0017088311981438863
+      "required_token_positions_per_second": 27832.636802880937,
+      "required_pages_per_second_upper_estimate": 0.0016948615547356717
     },
     {
       "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": 28062.043157730517,
-      "required_pages_per_second_upper_estimate": 0.0021602937629893425
+      "required_token_positions_per_second": 27832.636802880937,
+      "required_pages_per_second_upper_estimate": 0.002142633426755587
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28062.043157730517,
-      "required_pages_per_second_upper_estimate": 0.3484366584021073
+      "required_token_positions_per_second": 27832.636802880937,
+      "required_pages_per_second_upper_estimate": 0.34558819924855577
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13616705397640771,
+    "producer_batch_seconds": 0.10700062938292376,
     "producer_p95_seconds": 0.36575619599898346,
-    "gpu_update_seconds": 0.9852454379246988,
-    "producer_batches_per_second": 7.343920359569943,
-    "gpu_batches_per_second": 1.0149755193044891,
-    "producer_headroom": 7.2355640307486,
+    "gpu_update_seconds": 0.9933661763997212,
+    "producer_batches_per_second": 9.345739420104664,
+    "gpu_batches_per_second": 1.0066781251042005,
+    "producer_headroom": 9.283741433377518,
     "sustainable": true,
-    "buffer_seconds": 60.099971713406624,
+    "buffer_seconds": 60.595336760383,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310309.28186797595,
-    "upload_bytes_per_second": 5973278.84412117,
+    "generation_bytes_per_second": 324248.47729707306,
+    "upload_bytes_per_second": 5912657.1422590185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.58514825849016,
+    "estimated_full_shards_per_hour": 69.57617510971207,
     "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": 0.9471035349997692,
-      "bytes": 5896771,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.970585907001805,
       "bytes": 1087871,
@@ -562,6 +556,12 @@
       "bytes": 7782602,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8252346610024688,
+      "bytes": 897645,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -272,3 +272,4 @@
 {"step": 112200, "loss": 0.5894890539348125, "lr": 0.0003, "grad_norm": 0.13322304199477333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16418162723, "tokens_per_second": 147506.0320255269, "data_wait_s": 9.723997209221125e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 112300, "loss": 0.5872876457870007, "lr": 0.0003, "grad_norm": 0.14929047942423196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16432892142, "tokens_per_second": 149901.6128204821, "data_wait_s": 1.7496997315902263e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112400, "loss": 0.6337862759828568, "lr": 0.0003, "grad_norm": 0.17847442054156323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16447657916, "tokens_per_second": 150064.66074569745, "data_wait_s": 1.2643999070860445e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 112500, "loss": 0.5730115100741386, "lr": 0.0003, "grad_norm": 0.12394509323280861, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16462338797, "tokens_per_second": 133919.43582881003, "data_wait_s": 1.099500514101237e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:575414086846eb9b110f81ae96e919a7922a6b7bd9519796dd03581f7022bd76
-size 584412
+oid sha256:bbfcb1102ae7aa3d9f7ae6655755d2457274da4ecf7f726f0f16646495df7294
+size 586530