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": 75029.79867384241,
-      "required_pages_per_second_upper_estimate": 0.00650339278776246
+      "required_token_positions_per_second": 75192.39202405035,
+      "required_pages_per_second_upper_estimate": 0.006517485967269406
     },
     {
       "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": 9378.724834230301,
-      "required_pages_per_second_upper_estimate": 0.003741313560806726
+      "required_token_positions_per_second": 9399.049003006294,
+      "required_pages_per_second_upper_estimate": 0.0037494211756048724
     },
     {
       "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": 18757.449668460602,
-      "required_pages_per_second_upper_estimate": 0.004339205860959188
+      "required_token_positions_per_second": 18798.098006012588,
+      "required_pages_per_second_upper_estimate": 0.004348609138465532
     },
     {
       "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": 28136.174502690905,
-      "required_pages_per_second_upper_estimate": 0.0017133454081148655
+      "required_token_positions_per_second": 28197.147009018878,
+      "required_pages_per_second_upper_estimate": 0.0017170583138522214
     },
     {
       "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": 28136.174502690905,
-      "required_pages_per_second_upper_estimate": 0.00216600059913309
+      "required_token_positions_per_second": 28197.147009018878,
+      "required_pages_per_second_upper_estimate": 0.002170694431452917
     },
     {
       "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": 28136.174502690905,
-      "required_pages_per_second_upper_estimate": 0.34935712160486365
+      "required_token_positions_per_second": 28197.147009018878,
+      "required_pages_per_second_upper_estimate": 0.3501141960716053
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.34714176647634076,
-    "producer_p95_seconds": 0.38604618499812204,
-    "gpu_update_seconds": 0.9826495779430066,
-    "producer_batches_per_second": 2.8806674868037074,
-    "gpu_batches_per_second": 1.0176567745475589,
-    "producer_headroom": 2.8306866901018046,
+    "producer_batch_seconds": 0.1071812924217852,
+    "producer_p95_seconds": 0.4027518160000909,
+    "gpu_update_seconds": 0.9805247314970116,
+    "producer_batches_per_second": 9.329986394125104,
+    "gpu_batches_per_second": 1.0198620879998148,
+    "producer_headroom": 9.148282403970288,
     "sustainable": true,
-    "buffer_seconds": 60.924273832466405,
+    "buffer_seconds": 60.792533352814715,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312654.2696739578,
-    "upload_bytes_per_second": 5408427.460491201,
+    "generation_bytes_per_second": 336651.9501857303,
+    "upload_bytes_per_second": 5422351.308141541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.08832805313158,
+    "estimated_full_shards_per_hour": 72.23767165354664,
     "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": 0.7573466569992888,
-      "bytes": 999325,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.4190698960010195,
       "bytes": 13694673,
@@ -562,6 +556,12 @@
       "bytes": 1220951,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 2.1309150849992875,
+      "bytes": 9524825,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -394,3 +394,4 @@
 {"step": 354400, "loss": 0.5938422046601772, "lr": 0.0003, "grad_norm": 0.16056053732413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51874410027, "tokens_per_second": 148326.8045222818, "data_wait_s": 1.3731994840782136e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354500, "loss": 0.6181938368827105, "lr": 0.0003, "grad_norm": 0.18626586511146814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51889043122, "tokens_per_second": 148291.947073688, "data_wait_s": 1.5765006537549198e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354600, "loss": 0.5678000226616859, "lr": 0.0003, "grad_norm": 0.1558815531563137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51903795748, "tokens_per_second": 149932.9615391219, "data_wait_s": 1.1664997145999223e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 354700, "loss": 0.56468465924263, "lr": 0.0003, "grad_norm": 0.16401846710138815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51918300294, "tokens_per_second": 146860.8932387149, "data_wait_s": 1.1041003745049238e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cdaf8f27307ccd3e528ae81fe236465d808e82fa8accfcf0105112a0668c9fc8
-size 844374
+oid sha256:bf8d14f9e13c0ae62818225bebc600bd7c61bdcebb40d7f698bbaf7570e50cb5
+size 846492