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": 75775.5133751127,
-      "required_pages_per_second_upper_estimate": 0.006568029448071938
+      "required_token_positions_per_second": 75557.78675616026,
+      "required_pages_per_second_upper_estimate": 0.006549157456564211
     },
     {
       "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": 9471.939171889087,
-      "required_pages_per_second_upper_estimate": 0.0037784981537773606
+      "required_token_positions_per_second": 9444.723344520033,
+      "required_pages_per_second_upper_estimate": 0.0037676413533269636
     },
     {
       "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": 18943.878343778175,
-      "required_pages_per_second_upper_estimate": 0.004382332853961284
+      "required_token_positions_per_second": 18889.446689040065,
+      "required_pages_per_second_upper_estimate": 0.004369741048602027
     },
     {
       "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": 28415.817515667266,
-      "required_pages_per_second_upper_estimate": 0.001730374200431626
+      "required_token_positions_per_second": 28334.170033560098,
+      "required_pages_per_second_upper_estimate": 0.001725402297142532
     },
     {
       "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": 28415.817515667266,
-      "required_pages_per_second_upper_estimate": 0.0021875282923734148
+      "required_token_positions_per_second": 28334.170033560098,
+      "required_pages_per_second_upper_estimate": 0.002181242843186109
     },
     {
       "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": 28415.817515667266,
-      "required_pages_per_second_upper_estimate": 0.3528293519210706
+      "required_token_positions_per_second": 28334.170033560098,
+      "required_pages_per_second_upper_estimate": 0.35181556344984416
     }
   ],
   "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": 0.10065004402378008,
-    "producer_p95_seconds": 0.3895403419955983,
-    "gpu_update_seconds": 0.972979221335303,
-    "producer_batches_per_second": 9.935415425786948,
-    "gpu_batches_per_second": 1.0277711775053264,
-    "producer_headroom": 9.666952764624941,
+    "producer_batch_seconds": 0.15770345828093468,
+    "producer_p95_seconds": 0.3852658350006095,
+    "gpu_update_seconds": 0.9757829492535913,
+    "producer_batches_per_second": 6.341015034804049,
+    "gpu_batches_per_second": 1.0248180712369828,
+    "producer_headroom": 6.18745435192246,
     "sustainable": true,
-    "buffer_seconds": 60.324711722788784,
+    "buffer_seconds": 60.49854285372266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441345.37508234987,
-    "upload_bytes_per_second": 5325258.499660366,
+    "generation_bytes_per_second": 437742.2871861525,
+    "upload_bytes_per_second": 5253281.469810717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.702443498162,
+    "estimated_full_shards_per_hour": 93.92930471123152,
     "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": 1.134901017001539,
-      "bytes": 6770130,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.1226096480022534,
       "bytes": 7935280,
@@ -562,6 +556,12 @@
       "bytes": 6216009,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.4164216759963892,
+      "bytes": 2450413,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -331,3 +331,4 @@
 {"step": 148100, "loss": 0.6290670800954103, "lr": 0.0003, "grad_norm": 0.1458833417856235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21673815971, "tokens_per_second": 149335.9806593084, "data_wait_s": 9.645998943597078e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 148200, "loss": 0.5921144485473633, "lr": 0.0003, "grad_norm": 0.13957190930996768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21688248588, "tokens_per_second": 147450.79628526157, "data_wait_s": 1.4210003428161144e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 148300, "loss": 0.5635866448283195, "lr": 0.0003, "grad_norm": 0.15149851392395916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21702826596, "tokens_per_second": 149006.41855917856, "data_wait_s": 1.241300196852535e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 148400, "loss": 0.6222486458718777, "lr": 0.0003, "grad_norm": 0.1392784178267337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21717515881, "tokens_per_second": 150904.27421738647, "data_wait_s": 1.5240999346133322e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f6b8f22f3a730bea5b228b7195daa7b59696d941d382892f29974715a48cf43
-size 710157
+oid sha256:b5dc018e5e68fbc1448296bfee9cc0825abc01805ff34f9755454fc1aa0c0adb
+size 712275