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": 77041.54024260661,
-      "required_pages_per_second_upper_estimate": 0.005329912514985348
+      "required_token_positions_per_second": 76994.46151829694,
+      "required_pages_per_second_upper_estimate": 0.005326655499599915
     },
     {
       "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": 9630.192530325827,
-      "required_pages_per_second_upper_estimate": 0.004031960511159401
+      "required_token_positions_per_second": 9624.307689787118,
+      "required_pages_per_second_upper_estimate": 0.004029496651315288
     },
     {
       "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": 19260.385060651653,
-      "required_pages_per_second_upper_estimate": 0.004011430277622843
+      "required_token_positions_per_second": 19248.615379574236,
+      "required_pages_per_second_upper_estimate": 0.004008978963441779
     },
     {
       "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": 28890.577590977482,
-      "required_pages_per_second_upper_estimate": 0.0017319460132213648
+      "required_token_positions_per_second": 28872.923069361354,
+      "required_pages_per_second_upper_estimate": 0.0017308876516073715
     },
     {
       "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": 28890.577590977482,
-      "required_pages_per_second_upper_estimate": 0.0022461153314775055
+      "required_token_positions_per_second": 28872.923069361354,
+      "required_pages_per_second_upper_estimate": 0.0022447427700758855
     },
     {
       "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": 28890.577590977482,
-      "required_pages_per_second_upper_estimate": 0.370577309051673
+      "required_token_positions_per_second": 28872.923069361354,
+      "required_pages_per_second_upper_estimate": 0.37035085580432975
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3272097923670856,
-    "producer_p95_seconds": 0.3086001359988586,
-    "gpu_update_seconds": 0.95699021291407,
-    "producer_batches_per_second": 3.0561432552670484,
-    "gpu_batches_per_second": 1.0449427658773685,
-    "producer_headroom": 2.924699184553912,
+    "producer_batch_seconds": 0.12944419024248077,
+    "producer_p95_seconds": 0.2932223790012358,
+    "gpu_update_seconds": 0.9575753703073733,
+    "producer_batches_per_second": 7.725337059367085,
+    "gpu_batches_per_second": 1.0443042198119703,
+    "producer_headroom": 7.397592495372711,
     "sustainable": true,
-    "buffer_seconds": 60.29038341358641,
+    "buffer_seconds": 60.327248329364515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371626.40340904077,
-    "upload_bytes_per_second": 5383529.083250419,
+    "generation_bytes_per_second": 385524.6588146341,
+    "upload_bytes_per_second": 5383029.600886166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.7423751516668,
+    "estimated_full_shards_per_hour": 82.72461722687976,
     "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": 1.3023044059991662,
-      "bytes": 5788392,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.0160050430022238,
       "bytes": 1159604,
@@ -562,6 +556,12 @@
       "bytes": 11541986,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.6100285119973705,
+      "bytes": 7404435,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -288,3 +288,4 @@
 {"step": 469000, "loss": 0.5352138876914978, "lr": 0.0003, "grad_norm": 0.1501758199924488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68647764717, "tokens_per_second": 150225.79737717827, "data_wait_s": 9.68199674389325e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 469100, "loss": 0.5794492959976196, "lr": 0.0003, "grad_norm": 0.14745706607623796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68662279231, "tokens_per_second": 150878.88829045263, "data_wait_s": 8.628998330095783e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 469200, "loss": 0.5728943347930908, "lr": 0.0003, "grad_norm": 0.14785872656051363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68676850819, "tokens_per_second": 150585.90830807685, "data_wait_s": 1.7526999727124348e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 469300, "loss": 0.5722887516021729, "lr": 0.0003, "grad_norm": 0.15890479592170773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68691498974, "tokens_per_second": 151630.33289074, "data_wait_s": 1.0546002158662304e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1053e9e42a51a40dba5aaf7bbe068160f3ec40d7b37349dab96c923746b414ca
-size 618300
+oid sha256:7bab0f83a3c06b3098167349662b78d2c0141ff78c035b46f5dc38df08770ea8
+size 620418