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": 76762.43084709403,
-      "required_pages_per_second_upper_estimate": 0.005310603079380765
+      "required_token_positions_per_second": 76356.34488543836,
+      "required_pages_per_second_upper_estimate": 0.0052825091103041785
     },
     {
       "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": 9595.303855886754,
-      "required_pages_per_second_upper_estimate": 0.00401735335173013
+      "required_token_positions_per_second": 9544.543110679795,
+      "required_pages_per_second_upper_estimate": 0.003996100887716874
     },
     {
       "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": 19190.607711773508,
-      "required_pages_per_second_upper_estimate": 0.003996897496003959
+      "required_token_positions_per_second": 19089.08622135959,
+      "required_pages_per_second_upper_estimate": 0.003975753246852483
     },
     {
       "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": 28785.91156766026,
-      "required_pages_per_second_upper_estimate": 0.0017256714449392615
+      "required_token_positions_per_second": 28633.629332039385,
+      "required_pages_per_second_upper_estimate": 0.0017165423574353003
     },
     {
       "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": 28785.91156766026,
-      "required_pages_per_second_upper_estimate": 0.002237978008541774
+      "required_token_positions_per_second": 28633.629332039385,
+      "required_pages_per_second_upper_estimate": 0.002226138734541017
     },
     {
       "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": 28785.91156766026,
-      "required_pages_per_second_upper_estimate": 0.36923476568617974
+      "required_token_positions_per_second": 28633.629332039385,
+      "required_pages_per_second_upper_estimate": 0.36728145267556067
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1040741355938053,
-    "producer_p95_seconds": 0.2695702289965993,
-    "gpu_update_seconds": 0.9604698442505237,
-    "producer_batches_per_second": 9.608535245518985,
-    "gpu_batches_per_second": 1.0411571024182675,
-    "producer_headroom": 9.228708350739286,
+    "producer_batch_seconds": 0.35945784135157055,
+    "producer_p95_seconds": 0.27627472600215697,
+    "gpu_update_seconds": 0.965577911182341,
+    "producer_batches_per_second": 2.7819674102531042,
+    "gpu_batches_per_second": 1.0356492090581375,
+    "producer_headroom": 2.6862062809695395,
     "sustainable": true,
-    "buffer_seconds": 60.50960018778299,
+    "buffer_seconds": 60.83140840448748,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428238.14238715783,
-    "upload_bytes_per_second": 4830858.618658852,
+    "generation_bytes_per_second": 401233.73666235484,
+    "upload_bytes_per_second": 4870733.456577512,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.88993648253489,
+    "estimated_full_shards_per_hour": 86.09541964438424,
     "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": 1.008943750999606,
-      "bytes": 1209380,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.5094498580001527,
       "bytes": 5827665,
@@ -562,6 +556,12 @@
       "bytes": 6391793,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.537398492997454,
+      "bytes": 7291338,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 909500, "loss": 0.5202333331108093, "lr": 3e-05, "grad_norm": 0.3383423403843563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133124866204, "tokens_per_second": 153046.82829895406, "data_wait_s": 1.6049001715146005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909600, "loss": 0.4660811722278595, "lr": 3e-05, "grad_norm": 0.3591909206618709, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133139673548, "tokens_per_second": 153915.0225433402, "data_wait_s": 1.711399818304926e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909700, "loss": 0.5299293994903564, "lr": 3e-05, "grad_norm": 0.3244476383491736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133154207567, "tokens_per_second": 150707.41689402473, "data_wait_s": 1.4583998563466594e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 909800, "loss": 0.5249332785606384, "lr": 3e-05, "grad_norm": 0.3468487905408248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133168760721, "tokens_per_second": 151445.54722398162, "data_wait_s": 1.8576000002212822e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6f400640f9d7cb4b143b76b2c90617796602c039afc16f501c729024eb3fb11
-size 526443
+oid sha256:7a2b4786fb8acebfa7f9c0f02e2c99e5494509a0b5f8113f7e51ae3ec622ec21
+size 528561