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": 76315.31528464453,
-      "required_pages_per_second_upper_estimate": 0.005279670587319471
+      "required_token_positions_per_second": 76180.54998880118,
+      "required_pages_per_second_upper_estimate": 0.005270347211454461
     },
     {
       "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": 9539.414410580566,
-      "required_pages_per_second_upper_estimate": 0.003993953608084764
+      "required_token_positions_per_second": 9522.568748600148,
+      "required_pages_per_second_upper_estimate": 0.003986900681190986
     },
     {
       "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": 19078.828821161133,
-      "required_pages_per_second_upper_estimate": 0.003973616900896979
+      "required_token_positions_per_second": 19045.137497200296,
+      "required_pages_per_second_upper_estimate": 0.003966599886615898
     },
     {
       "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": 28618.243231741697,
-      "required_pages_per_second_upper_estimate": 0.0017156199842156689
+      "required_token_positions_per_second": 28567.706245800444,
+      "required_pages_per_second_upper_estimate": 0.001712590369074
     },
     {
       "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": 28618.243231741697,
-      "required_pages_per_second_upper_estimate": 0.0022249425329191747
+      "required_token_positions_per_second": 28567.706245800444,
+      "required_pages_per_second_upper_estimate": 0.002221013504550951
     },
     {
       "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": 28618.243231741697,
-      "required_pages_per_second_upper_estimate": 0.36708409630124933
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28567.706245800444,
+      "required_pages_per_second_upper_estimate": 0.36643586210798273
     }
   ],
   "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": 0.4152253710546603,
-    "producer_p95_seconds": 0.25926168600017263,
-    "gpu_update_seconds": 0.9660970373378629,
-    "producer_batches_per_second": 2.4083306794573494,
-    "gpu_batches_per_second": 1.0350927094813982,
-    "producer_headroom": 2.3266811343536276,
+    "producer_batch_seconds": 0.08088264356254626,
+    "producer_p95_seconds": 0.21894708299987542,
+    "gpu_update_seconds": 0.9678060871290413,
+    "producer_batches_per_second": 12.363591939557509,
+    "gpu_batches_per_second": 1.0332648381727592,
+    "producer_headroom": 11.965559537883307,
     "sustainable": true,
-    "buffer_seconds": 60.864113352285365,
+    "buffer_seconds": 60.00397740200056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457932.35297836777,
-    "upload_bytes_per_second": 5667296.694447435,
+    "generation_bytes_per_second": 433673.9531381357,
+    "upload_bytes_per_second": 5672089.532033722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.26162282956385,
+    "estimated_full_shards_per_hour": 93.05633493049673,
     "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": 1.0933929989987519,
-      "bytes": 5793673,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0935950399998546,
       "bytes": 4381259,
@@ -562,6 +556,12 @@
       "bytes": 8524201,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.0446712169996317,
+      "bytes": 5880079,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -157,3 +157,4 @@
 {"step": 412800, "loss": 0.5927441716194153, "lr": 0.0003, "grad_norm": 0.17216612791240354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60423269331, "tokens_per_second": 151040.5523618154, "data_wait_s": 1.3699000191991217e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412900, "loss": 0.6292275190353394, "lr": 0.0003, "grad_norm": 0.16825339119515065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60437878206, "tokens_per_second": 150782.78168671072, "data_wait_s": 9.114999556913972e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 413000, "loss": 0.5962061882019043, "lr": 0.0003, "grad_norm": 0.1740489759002041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60452450337, "tokens_per_second": 151370.53684027633, "data_wait_s": 1.420499938831199e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 413100, "loss": 0.581549882888794, "lr": 0.0003, "grad_norm": 0.16567601139019777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60467029185, "tokens_per_second": 150324.64960564225, "data_wait_s": 1.538399919809308e-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 sha256:389660a5180d6fbab6e37bc6a6948938548b268c717afe05d36b38464e5affc9
-size 339276
+oid sha256:573ec3d6f51eda42c3e557a4ff3c8373ecdc3184a3a65f43c47e41ac124c406d
+size 341394