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": 77430.37409841757,
-      "required_pages_per_second_upper_estimate": 0.005356812943349197
+      "required_token_positions_per_second": 76537.58058976155,
+      "required_pages_per_second_upper_estimate": 0.005295047416854028
     },
     {
       "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": 9678.796762302196,
-      "required_pages_per_second_upper_estimate": 0.004052310088116126
+      "required_token_positions_per_second": 9567.197573720194,
+      "required_pages_per_second_upper_estimate": 0.004005585838312905
     },
     {
       "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": 19357.593524604392,
-      "required_pages_per_second_upper_estimate": 0.004031676237104612
+      "required_token_positions_per_second": 19134.395147440388,
+      "required_pages_per_second_upper_estimate": 0.003985189901278381
     },
     {
       "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": 29036.39028690659,
-      "required_pages_per_second_upper_estimate": 0.0017406872616992197
+      "required_token_positions_per_second": 28701.592721160578,
+      "required_pages_per_second_upper_estimate": 0.0017206166588390292
     },
     {
       "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": 29036.39028690659,
-      "required_pages_per_second_upper_estimate": 0.0022574516272237254
+      "required_token_positions_per_second": 28701.592721160578,
+      "required_pages_per_second_upper_estimate": 0.002231422589105834
     },
     {
       "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": 29036.39028690659,
-      "required_pages_per_second_upper_estimate": 0.37244763775357664
+      "required_token_positions_per_second": 28701.592721160578,
+      "required_pages_per_second_upper_estimate": 0.36815321405780554
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10163664038286413,
-    "producer_p95_seconds": 0.19971602099940355,
-    "gpu_update_seconds": 0.9521844735799457,
-    "producer_batches_per_second": 9.838971420474062,
-    "gpu_batches_per_second": 1.0502166625761933,
-    "producer_headroom": 9.368515822572224,
+    "producer_batch_seconds": 0.06261474749226181,
+    "producer_p95_seconds": 0.1975450110003294,
+    "gpu_update_seconds": 0.9632914893819183,
+    "producer_batches_per_second": 15.970678475124156,
+    "gpu_batches_per_second": 1.0381073756206807,
+    "producer_headroom": 15.384418654742094,
     "sustainable": true,
-    "buffer_seconds": 60.93980630911653,
+    "buffer_seconds": 60.68736383106086,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 529811.9739143956,
-    "upload_bytes_per_second": 5386298.614575342,
+    "generation_bytes_per_second": 529608.8085411245,
+    "upload_bytes_per_second": 5389253.883969207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.68531621049786,
+    "estimated_full_shards_per_hour": 113.64172165084172,
     "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": 0.9808759570000802,
-      "bytes": 5815079,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.6121151610000197,
       "bytes": 12779121,
@@ -562,6 +556,12 @@
       "bytes": 5891615,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2515704560009908,
+      "bytes": 7512167,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -100,3 +100,4 @@
 {"step": 895000, "loss": 0.547240674495697, "lr": 3e-05, "grad_norm": 0.4042976704834889, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131002760332, "tokens_per_second": 152778.88040975755, "data_wait_s": 1.3955001122667454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895100, "loss": 0.49131640791893005, "lr": 3e-05, "grad_norm": 0.3079410104055874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131017429681, "tokens_per_second": 142054.74784761734, "data_wait_s": 9.436000254936516e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 895200, "loss": 0.5449494123458862, "lr": 3e-05, "grad_norm": 0.315098659898938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131032076146, "tokens_per_second": 152288.34958202927, "data_wait_s": 1.6501999198226258e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 895300, "loss": 0.5230830907821655, "lr": 3e-05, "grad_norm": 0.3115013720633862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131046724901, "tokens_per_second": 153471.71832988693, "data_wait_s": 3.731300057552289e-05, "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 sha256:85a86c1b7749696556750ac974fa8ffcdfbc2ee082440c5da3c390e886923a8f
-size 217767
+oid sha256:10f924fddd230cecac2a13892a3d241fe0c8c0d951cc90f273fa2e24d50e6a31
+size 219885