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": 77137.54766905162,
-      "required_pages_per_second_upper_estimate": 0.005336554531514734
+      "required_token_positions_per_second": 77549.35840581302,
+      "required_pages_per_second_upper_estimate": 0.005365044553816444
     },
     {
       "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": 9642.193458631453,
-      "required_pages_per_second_upper_estimate": 0.004036985049233085
+      "required_token_positions_per_second": 9693.669800726628,
+      "required_pages_per_second_upper_estimate": 0.004058537118720076
     },
     {
       "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": 19284.386917262906,
-      "required_pages_per_second_upper_estimate": 0.004016429231383442
+      "required_token_positions_per_second": 19387.339601453255,
+      "required_pages_per_second_upper_estimate": 0.004037871560455431
     },
     {
       "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": 28926.58037589436,
-      "required_pages_per_second_upper_estimate": 0.001734104325204062
+      "required_token_positions_per_second": 29081.009402179883,
+      "required_pages_per_second_upper_estimate": 0.0017433621095304095
     },
     {
       "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": 28926.58037589436,
-      "required_pages_per_second_upper_estimate": 0.0022489143896452777
+      "required_token_positions_per_second": 29081.009402179883,
+      "required_pages_per_second_upper_estimate": 0.0022609205671774773
     },
     {
       "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": 28926.58037589436,
-      "required_pages_per_second_upper_estimate": 0.3710391141198081
+      "required_token_positions_per_second": 29081.009402179883,
+      "required_pages_per_second_upper_estimate": 0.3730199638560291
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.080782925539026,
-    "producer_p95_seconds": 0.17433709899978567,
-    "gpu_update_seconds": 0.9557991176530548,
-    "producer_batches_per_second": 12.378853493203868,
-    "gpu_batches_per_second": 1.0462449499383089,
-    "producer_headroom": 11.831697246360694,
+    "producer_batch_seconds": 0.07607324010155025,
+    "producer_p95_seconds": 0.17587845800062496,
+    "gpu_update_seconds": 0.9507235329296215,
+    "producer_batches_per_second": 13.145226871697576,
+    "gpu_batches_per_second": 1.0518304905302331,
+    "producer_headroom": 12.497476532621716,
     "sustainable": true,
-    "buffer_seconds": 60.21534441214245,
+    "buffer_seconds": 60.84630610749578,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320676.7694696612,
-    "upload_bytes_per_second": 5227695.851804509,
+    "generation_bytes_per_second": 359797.28737091454,
+    "upload_bytes_per_second": 5160677.06948717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.80976975505234,
+    "estimated_full_shards_per_hour": 77.204122217613,
     "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": [
@@ -382,6 +382,12 @@
       "bytes": 6044727,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0941710249999232,
+      "bytes": 2736868,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -54,3 +54,4 @@
 {"step": 890400, "loss": 0.5553457736968994, "lr": 3e-05, "grad_norm": 0.32650330970892855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130329770014, "tokens_per_second": 152760.38046780383, "data_wait_s": 1.0931999895547051e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890500, "loss": 0.5742422938346863, "lr": 3e-05, "grad_norm": 0.33076722199345765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130344441750, "tokens_per_second": 153408.29754627956, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890600, "loss": 0.5161193013191223, "lr": 3e-05, "grad_norm": 0.32479326522699076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130359127841, "tokens_per_second": 152914.48461358098, "data_wait_s": 1.3161999959265813e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 890700, "loss": 0.5632334351539612, "lr": 3e-05, "grad_norm": 0.31817407662024405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130373755907, "tokens_per_second": 142281.99469000564, "data_wait_s": 9.66199968388537e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c61c8548dc98f976630266989ef8dbaddbb50a9b5b8d15b66bc3de422441250
-size 119556
+oid sha256:773aaa73dc5b6b4fa41b2105b6fb2949c40955f228fcc641518986f02951fc82
+size 121674