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": 76756.4594248897,
-      "required_pages_per_second_upper_estimate": 0.005310189962537061
+      "required_token_positions_per_second": 75638.54061639706,
+      "required_pages_per_second_upper_estimate": 0.0052328497454365306
     },
     {
       "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": 9594.557428111213,
-      "required_pages_per_second_upper_estimate": 0.004017040838007696
+      "required_token_positions_per_second": 9454.817577049633,
+      "required_pages_per_second_upper_estimate": 0.00395853468046813
     },
     {
       "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": 19189.114856222426,
-      "required_pages_per_second_upper_estimate": 0.0039965865735619175
+      "required_token_positions_per_second": 18909.635154099265,
+      "required_pages_per_second_upper_estimate": 0.003938378321985045
     },
     {
       "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": 28783.67228433364,
-      "required_pages_per_second_upper_estimate": 0.0017255372033230182
+      "required_token_positions_per_second": 28364.452731148896,
+      "required_pages_per_second_upper_estimate": 0.001700405631220787
     },
     {
       "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": 28783.67228433364,
-      "required_pages_per_second_upper_estimate": 0.0022378039141126957
+      "required_token_positions_per_second": 28364.452731148896,
+      "required_pages_per_second_upper_estimate": 0.0022052114378045213
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28783.67228433364,
-      "required_pages_per_second_upper_estimate": 0.36920604256402095
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28364.452731148896,
+      "required_pages_per_second_upper_estimate": 0.36382874425865364
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09835028253897349,
-    "producer_p95_seconds": 0.27294528299898957,
-    "gpu_update_seconds": 0.9605445659221266,
-    "producer_batches_per_second": 10.167738965098831,
-    "gpu_batches_per_second": 1.0410761098210952,
-    "producer_headroom": 9.76656641064035,
+    "producer_batch_seconds": 0.13433134303136285,
+    "producer_p95_seconds": 0.29478266799924313,
+    "gpu_update_seconds": 0.9747411755855204,
+    "producer_batches_per_second": 7.444279029999173,
+    "gpu_batches_per_second": 1.0259133655652812,
+    "producer_headroom": 7.256245293088032,
     "sustainable": true,
-    "buffer_seconds": 60.51430765309397,
+    "buffer_seconds": 60.433952886302265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484422.1317484644,
-    "upload_bytes_per_second": 5932044.132380166,
+    "generation_bytes_per_second": 500820.9743037416,
+    "upload_bytes_per_second": 5897838.74281652,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.94571270313692,
+    "estimated_full_shards_per_hour": 107.46452256998239,
     "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": 1.0314974919965607,
-      "bytes": 5945434,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1649709810008062,
       "bytes": 7159686,
@@ -562,6 +556,12 @@
       "bytes": 6077154,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.8500619599981292,
+      "bytes": 2427201,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -266,3 +266,4 @@
 {"step": 423700, "loss": 0.5932236909866333, "lr": 0.0003, "grad_norm": 0.17331497094262296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62017961215, "tokens_per_second": 152020.00371867706, "data_wait_s": 2.2943000658415258e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423800, "loss": 0.5542709231376648, "lr": 0.0003, "grad_norm": 0.13283663005110638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62032678547, "tokens_per_second": 152169.60309610044, "data_wait_s": 1.1085998266935349e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423900, "loss": 0.5853983163833618, "lr": 0.0003, "grad_norm": 0.15177953842616437, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62047360879, "tokens_per_second": 151233.1366609056, "data_wait_s": 1.0456999007146806e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 424000, "loss": 0.6058544516563416, "lr": 0.0003, "grad_norm": 0.17613054731147765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62062117848, "tokens_per_second": 153140.9930948525, "data_wait_s": 1.4223001926438883e-05, "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 sha256:04f27ce974bbfd0df67b5f53a5d3698c0a9c3e3eb378ac57dd5f966939ac231b
-size 571704
+oid sha256:1e5ecbe35e5a463fa3d11c46d5cec85495f9de6055263b7a56929afb843a6189
+size 573822