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": 77770.21218168066,
-      "required_pages_per_second_upper_estimate": 0.005380323730482329
+      "required_token_positions_per_second": 76718.54820728535,
+      "required_pages_per_second_upper_estimate": 0.005307567176537048
     },
     {
       "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": 9721.276522710083,
-      "required_pages_per_second_upper_estimate": 0.004070095476720638
+      "required_token_positions_per_second": 9589.818525910669,
+      "required_pages_per_second_upper_estimate": 0.004015056758615859
     },
     {
       "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": 19442.553045420165,
-      "required_pages_per_second_upper_estimate": 0.004049371064757304
+      "required_token_positions_per_second": 19179.637051821337,
+      "required_pages_per_second_upper_estimate": 0.003994612596851681
     },
     {
       "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": 29163.829568130248,
-      "required_pages_per_second_upper_estimate": 0.0017483270520200626
+      "required_token_positions_per_second": 28769.455577732006,
+      "required_pages_per_second_upper_estimate": 0.0017246849334699042
     },
     {
       "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": 29163.829568130248,
-      "required_pages_per_second_upper_estimate": 0.0022673594707927074
+      "required_token_positions_per_second": 28769.455577732006,
+      "required_pages_per_second_upper_estimate": 0.002236698627707102
     },
     {
       "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": 29163.829568130248,
-      "required_pages_per_second_upper_estimate": 0.37408229201947446
+      "required_token_positions_per_second": 28769.455577732006,
+      "required_pages_per_second_upper_estimate": 0.36902368591644547
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07720620476570161,
-    "producer_p95_seconds": 0.18634220300009474,
-    "gpu_update_seconds": 0.9480236446798221,
-    "producer_batches_per_second": 12.952326863296923,
-    "gpu_batches_per_second": 1.0548260115787849,
-    "producer_headroom": 12.279112120027118,
+    "producer_batch_seconds": 0.10334832367971103,
+    "producer_p95_seconds": 0.18851638300111517,
+    "gpu_update_seconds": 0.9610192283721897,
+    "producer_batches_per_second": 9.67601567587222,
+    "gpu_batches_per_second": 1.0405619060232931,
+    "producer_headroom": 9.298837118543931,
     "sustainable": true,
-    "buffer_seconds": 60.67351325950862,
+    "buffer_seconds": 60.54421138744795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320450.026537929,
-    "upload_bytes_per_second": 5821490.627294556,
+    "generation_bytes_per_second": 250370.21790817063,
+    "upload_bytes_per_second": 5840838.470627583,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.76111600020793,
+    "estimated_full_shards_per_hour": 53.723620442713155,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 812365,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.1693048819997784,
+      "bytes": 7948420,
+      "pages": 20,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 491500, "loss": 0.664214551448822, "lr": 0.0003, "grad_norm": 0.1895744930643989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71939047513, "tokens_per_second": 154091.79369911706, "data_wait_s": 1.1963998986175284e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491600, "loss": 0.5676164627075195, "lr": 0.0003, "grad_norm": 0.1384545730428987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71953732155, "tokens_per_second": 153761.22089275956, "data_wait_s": 1.5396000890177675e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491700, "loss": 0.5966281294822693, "lr": 0.0003, "grad_norm": 0.16710601263201885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71968244192, "tokens_per_second": 78997.4518323443, "data_wait_s": 1.614299981156364e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 491800, "loss": 0.5912267565727234, "lr": 0.0003, "grad_norm": 0.17450495071352756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71982856035, "tokens_per_second": 153186.3163065317, "data_wait_s": 1.0395999197498895e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5268251d60d36cd46f71c1548252091a9fd3a33c4bcea846d7a3ec76caee942
-size 142854
+oid sha256:40c365d785fd9182fc550a176af47f94599378fbdf337827baff6bc1cc57faa9
+size 144972