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": 77155.77055166423,
-      "required_pages_per_second_upper_estimate": 0.005337815232816788
+      "required_token_positions_per_second": 76762.43084709403,
+      "required_pages_per_second_upper_estimate": 0.005310603079380765
     },
     {
       "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": 9644.471318958029,
-      "required_pages_per_second_upper_estimate": 0.0040379387417846906
+      "required_token_positions_per_second": 9595.303855886754,
+      "required_pages_per_second_upper_estimate": 0.00401735335173013
     },
     {
       "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": 19288.942637916058,
-      "required_pages_per_second_upper_estimate": 0.004017378067853061
+      "required_token_positions_per_second": 19190.607711773508,
+      "required_pages_per_second_upper_estimate": 0.003996897496003959
     },
     {
       "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": 28933.413956874087,
-      "required_pages_per_second_upper_estimate": 0.0017345139879495228
+      "required_token_positions_per_second": 28785.91156766026,
+      "required_pages_per_second_upper_estimate": 0.0017256714449392615
     },
     {
       "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": 28933.413956874087,
-      "required_pages_per_second_upper_estimate": 0.0022494456705087055
+      "required_token_positions_per_second": 28785.91156766026,
+      "required_pages_per_second_upper_estimate": 0.002237978008541774
     },
     {
       "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": 28933.413956874087,
-      "required_pages_per_second_upper_estimate": 0.37112676795928845
+      "required_token_positions_per_second": 28785.91156766026,
+      "required_pages_per_second_upper_estimate": 0.36923476568617974
     }
   ],
   "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": 0.09724721418749027,
-    "producer_p95_seconds": 0.28618997899684473,
-    "gpu_update_seconds": 0.955573374134486,
-    "producer_batches_per_second": 10.283070917301798,
-    "gpu_batches_per_second": 1.0464921136022167,
-    "producer_headroom": 9.826228772910282,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.20112257047261,
+    "buffer_seconds": 60.50960018778299,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412283.1753287663,
-    "upload_bytes_per_second": 4930776.978256061,
+    "generation_bytes_per_second": 428238.14238715783,
+    "upload_bytes_per_second": 4830858.618658852,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.46637196442835,
+    "estimated_full_shards_per_hour": 91.88993648253489,
     "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": 1.710491572997853,
-      "bytes": 17528313,
-      "pages": 40,
-      "shards": 5
-    },
     {
       "seconds": 1.008943750999606,
       "bytes": 1209380,
@@ -562,6 +556,12 @@
       "bytes": 7112834,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.2346669860016846,
+      "bytes": 6391793,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -244,3 +244,4 @@
 {"step": 909400, "loss": 0.500087559223175, "lr": 3e-05, "grad_norm": 0.31723700030761487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133110124186, "tokens_per_second": 153149.113347913, "data_wait_s": 1.7561000277055427e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:390d94f22c2318dec14706090ec5079d040c14f40d8ffd602e621c8021d17b9e
-size 524325
+oid sha256:a6f400640f9d7cb4b143b76b2c90617796602c039afc16f501c729024eb3fb11
+size 526443