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": 75721.4939418217,
-      "required_pages_per_second_upper_estimate": 0.006563347180505275
+      "required_token_positions_per_second": 75678.71474192127,
+      "required_pages_per_second_upper_estimate": 0.006559639187880798
     },
     {
       "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": 9465.186742727712,
-      "required_pages_per_second_upper_estimate": 0.0037758045088270753
+      "required_token_positions_per_second": 9459.839342740159,
+      "required_pages_per_second_upper_estimate": 0.0037736713510212855
     },
     {
       "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": 18930.373485455424,
-      "required_pages_per_second_upper_estimate": 0.004379208742665704
+      "required_token_positions_per_second": 18919.678685480318,
+      "required_pages_per_second_upper_estimate": 0.004376734688913514
     },
     {
       "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": 28395.560228183134,
-      "required_pages_per_second_upper_estimate": 0.001729140637905616
+      "required_token_positions_per_second": 28379.518028220475,
+      "required_pages_per_second_upper_estimate": 0.0017281637520948083
     },
     {
       "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": 28395.560228183134,
-      "required_pages_per_second_upper_estimate": 0.0021859688303071255
+      "required_token_positions_per_second": 28379.518028220475,
+      "required_pages_per_second_upper_estimate": 0.0021847338573463483
     },
     {
       "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": 28395.560228183134,
-      "required_pages_per_second_upper_estimate": 0.3525778242072977
+      "required_token_positions_per_second": 28379.518028220475,
+      "required_pages_per_second_upper_estimate": 0.352378633773551
     }
   ],
   "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": 0.1644502504530152,
-    "producer_p95_seconds": 0.21896746200036432,
-    "gpu_update_seconds": 0.9736733411077001,
-    "producer_batches_per_second": 6.080866385093821,
-    "gpu_batches_per_second": 1.027038492049448,
-    "producer_headroom": 5.920777490003803,
+    "producer_batch_seconds": 0.6163119400155068,
+    "producer_p95_seconds": 0.20353663399873767,
+    "gpu_update_seconds": 0.974223733204593,
+    "producer_batches_per_second": 1.622554967821716,
+    "gpu_batches_per_second": 1.0264582620160763,
+    "producer_headroom": 1.5807315580809305,
     "sustainable": true,
-    "buffer_seconds": 60.36774714867741,
+    "buffer_seconds": 60.40187145868477,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246784.7349944752,
-    "upload_bytes_per_second": 6537119.279500135,
+    "generation_bytes_per_second": 211616.29725585136,
+    "upload_bytes_per_second": 6587034.816231401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.954259275204585,
+    "estimated_full_shards_per_hour": 45.407931215826565,
     "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": 0.9600033409997195,
-      "bytes": 1092240,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 0.9772284549999313,
       "bytes": 8526326,
@@ -562,6 +556,12 @@
       "bytes": 8597889,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.1277135729997099,
+      "bytes": 5536865,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -116,3 +116,4 @@
 {"step": 11600, "loss": 0.6731462813913822, "lr": 0.0003, "grad_norm": 0.14444594314923076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1696884254, "tokens_per_second": 147854.69487109123, "data_wait_s": 1.1763999282266013e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11700, "loss": 0.7302270978689194, "lr": 0.0003, "grad_norm": 0.1457306650160281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1711427439, "tokens_per_second": 146573.4266490581, "data_wait_s": 1.314000110141933e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11800, "loss": 0.7083174623548985, "lr": 0.0003, "grad_norm": 0.15130096636100004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1725960527, "tokens_per_second": 148386.64885272944, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11900, "loss": 0.6477013938128948, "lr": 0.0003, "grad_norm": 0.1480410445894246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1740609258, "tokens_per_second": 110180.41730813945, "data_wait_s": 1.3979000868857838e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a62169484aeedb9535c4e68c02bcf31eaa53afdb304aa2da56ea8274daee8e6
-size 247466
+oid sha256:d485ca2ae7c74f948fe8eff2661f728b1123cf872464dc4853e06449560f9305
+size 249549