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": 76821.55630227453,
-      "required_pages_per_second_upper_estimate": 0.005314693515560109
+      "required_token_positions_per_second": 76595.39191360549,
+      "required_pages_per_second_upper_estimate": 0.005299046938378304
     },
     {
       "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": 9602.694537784317,
-      "required_pages_per_second_upper_estimate": 0.004020447675905652
+      "required_token_positions_per_second": 9574.423989200686,
+      "required_pages_per_second_upper_estimate": 0.004008611387569872
     },
     {
       "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": 19205.389075568633,
-      "required_pages_per_second_upper_estimate": 0.0039999760642717075
+      "required_token_positions_per_second": 19148.847978401373,
+      "required_pages_per_second_upper_estimate": 0.00398820004482077
     },
     {
       "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": 28808.08361335295,
-      "required_pages_per_second_upper_estimate": 0.0017270006252237847
+      "required_token_positions_per_second": 28723.271967602057,
+      "required_pages_per_second_upper_estimate": 0.0017219162965609034
     },
     {
       "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": 28808.08361335295,
-      "required_pages_per_second_upper_estimate": 0.002239701787569858
+      "required_token_positions_per_second": 28723.271967602057,
+      "required_pages_per_second_upper_estimate": 0.002233108055159731
     },
     {
       "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": 28808.08361335295,
-      "required_pages_per_second_upper_estimate": 0.36951916488183767
+      "required_token_positions_per_second": 28723.271967602057,
+      "required_pages_per_second_upper_estimate": 0.36843129215379555
     }
   ],
   "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": 0.10970381251556205,
-    "producer_p95_seconds": 0.28652801100179204,
-    "gpu_update_seconds": 0.9597306218309074,
-    "producer_batches_per_second": 9.11545348397208,
-    "gpu_batches_per_second": 1.041959042728333,
-    "producer_headroom": 8.748379840443237,
+    "producer_batch_seconds": 0.1331195171405568,
+    "producer_p95_seconds": 0.2928424299971084,
+    "gpu_update_seconds": 0.9625644331601604,
+    "producer_batches_per_second": 7.512046478835487,
+    "gpu_batches_per_second": 1.038891491883755,
+    "producer_headroom": 7.2308287607730595,
     "sustainable": true,
-    "buffer_seconds": 60.46302917534717,
+    "buffer_seconds": 60.641559289090104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322302.024100579,
-    "upload_bytes_per_second": 5898838.355816172,
+    "generation_bytes_per_second": 337235.2815844054,
+    "upload_bytes_per_second": 5919229.313440888,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.15851156485583,
+    "estimated_full_shards_per_hour": 72.36284099244233,
     "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": 1.03145402500013,
-      "bytes": 6170934,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9970344739995198,
       "bytes": 6228444,
@@ -562,6 +556,12 @@
       "bytes": 6207933,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9747246619990619,
+      "bytes": 7297014,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 427600, "loss": 0.5873361229896545, "lr": 0.0003, "grad_norm": 0.17662522692257182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62588560943, "tokens_per_second": 152404.64430579793, "data_wait_s": 1.518800127087161e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427700, "loss": 0.6278319954872131, "lr": 0.0003, "grad_norm": 0.14338572818608436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62603273646, "tokens_per_second": 152269.30883924387, "data_wait_s": 8.983999578049406e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 427800, "loss": 0.5700190663337708, "lr": 0.0003, "grad_norm": 0.1584146694789872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62617848647, "tokens_per_second": 150644.28702120512, "data_wait_s": 1.571300163050182e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 427900, "loss": 0.5963143706321716, "lr": 0.0003, "grad_norm": 0.14141060153501045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62632384124, "tokens_per_second": 120423.55888231561, "data_wait_s": 7.5739990279544145e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7335b7f8bd084f85430f51748f5ce590e8c256ce03ecf89c1bbd616ab8980c8e
-size 655089
+oid sha256:ac106cf3f5d12a087b6b37460fb76f66255644c75583ab8a0a8ea32618e1bacc
+size 657207