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": 75456.05175386052,
-      "required_pages_per_second_upper_estimate": 0.006540339324409876
+      "required_token_positions_per_second": 75516.44573878514,
+      "required_pages_per_second_upper_estimate": 0.006545574121955986
     },
     {
       "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": 9432.006469232565,
-      "required_pages_per_second_upper_estimate": 0.0037625684016405637
+      "required_token_positions_per_second": 9439.555717348143,
+      "required_pages_per_second_upper_estimate": 0.0037655799095851856
     },
     {
       "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": 18864.01293846513,
-      "required_pages_per_second_upper_estimate": 0.004363857398025238
+      "required_token_positions_per_second": 18879.111434696286,
+      "required_pages_per_second_upper_estimate": 0.004367350169403858
     },
     {
       "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": 28296.019407697695,
-      "required_pages_per_second_upper_estimate": 0.0017230791241883761
+      "required_token_positions_per_second": 28318.66715204443,
+      "required_pages_per_second_upper_estimate": 0.0017244582529955608
     },
     {
       "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": 28296.019407697695,
-      "required_pages_per_second_upper_estimate": 0.0021783059023995294
+      "required_token_positions_per_second": 28318.66715204443,
+      "required_pages_per_second_upper_estimate": 0.0021800493884523097
     },
     {
       "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": 28296.019407697695,
-      "required_pages_per_second_upper_estimate": 0.35134186035856435
+      "required_token_positions_per_second": 28318.66715204443,
+      "required_pages_per_second_upper_estimate": 0.3516230695462263
     }
   ],
   "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": 0.12182848796857115,
-    "producer_p95_seconds": 0.39066915299918037,
-    "gpu_update_seconds": 0.9770985664675715,
-    "producer_batches_per_second": 8.20826078263383,
-    "gpu_batches_per_second": 1.0234382019566586,
-    "producer_headroom": 8.020279843903502,
+    "producer_batch_seconds": 0.151782601375146,
+    "producer_p95_seconds": 0.4028094749955926,
+    "gpu_update_seconds": 0.9763171356743741,
+    "producer_batches_per_second": 6.5883704122872375,
+    "gpu_batches_per_second": 1.0242573478025327,
+    "producer_headroom": 6.432338929686071,
     "sustainable": true,
-    "buffer_seconds": 60.580111120989436,
+    "buffer_seconds": 60.53166241181119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306839.9424375289,
-    "upload_bytes_per_second": 5956251.059347451,
+    "generation_bytes_per_second": 320074.4784073047,
+    "upload_bytes_per_second": 5906805.640999986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.84070877880478,
+    "estimated_full_shards_per_hour": 68.680532113689,
     "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": 0.9876665679985308,
-      "bytes": 5790107,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.9691603890023543,
       "bytes": 2643836,
@@ -562,6 +556,12 @@
       "bytes": 10382512,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9968646140041528,
+      "bytes": 2322828,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -339,3 +339,4 @@
 {"step": 118900, "loss": 0.5921273212879896, "lr": 0.0003, "grad_norm": 0.13395452390409357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17399398353, "tokens_per_second": 143621.63730805874, "data_wait_s": 1.3270997442305088e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119000, "loss": 0.6057633049786091, "lr": 0.0003, "grad_norm": 0.15477933710314376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17414016155, "tokens_per_second": 148428.89229204145, "data_wait_s": 1.7439000657759607e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119100, "loss": 0.5712260492146015, "lr": 0.0003, "grad_norm": 0.13741100760686165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17428635994, "tokens_per_second": 148904.5290307736, "data_wait_s": 1.4810997527092695e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 119200, "loss": 0.5859232051298022, "lr": 0.0003, "grad_norm": 0.13938095464892203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17443251462, "tokens_per_second": 148488.68158974353, "data_wait_s": 1.6275000234600157e-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:bebb5fb09e0cde40b9c07261f1ec132a30826f06767c4845bcb060bf8d931835
-size 727101
+oid sha256:9e69bf880e56c6f68f28a21a409923286e57215bbec2ba073b60a0fc9c1cce1b
+size 729219