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": 76384.36451691357,
-      "required_pages_per_second_upper_estimate": 0.005284447573424136
+      "required_token_positions_per_second": 76157.50611190939,
+      "required_pages_per_second_upper_estimate": 0.0052687529825819245
     },
     {
       "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": 9548.045564614196,
-      "required_pages_per_second_upper_estimate": 0.0039975672920396525
+      "required_token_positions_per_second": 9519.688263988673,
+      "required_pages_per_second_upper_estimate": 0.003985694682435521
     },
     {
       "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": 19096.091129228393,
-      "required_pages_per_second_upper_estimate": 0.003977212184429712
+      "required_token_positions_per_second": 19039.376527977347,
+      "required_pages_per_second_upper_estimate": 0.003965400028653733
     },
     {
       "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": 28644.136693842593,
-      "required_pages_per_second_upper_estimate": 0.00171717225773159
+      "required_token_positions_per_second": 28559.064791966022,
+      "required_pages_per_second_upper_estimate": 0.0017120723271113625
     },
     {
       "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": 28644.136693842593,
-      "required_pages_per_second_upper_estimate": 0.002226955635704216
+      "required_token_positions_per_second": 28559.064791966022,
+      "required_pages_per_second_upper_estimate": 0.0022203416695250632
     },
     {
       "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": 28644.136693842593,
-      "required_pages_per_second_upper_estimate": 0.3674162298308461
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28559.064791966022,
+      "required_pages_per_second_upper_estimate": 0.366325018816665
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07670598343744928,
-    "producer_p95_seconds": 0.1814573969995763,
-    "gpu_update_seconds": 0.9652237138619464,
-    "producer_batches_per_second": 13.03679263581127,
-    "gpu_batches_per_second": 1.0360292496326169,
-    "producer_headroom": 12.583421404785827,
+    "producer_batch_seconds": 0.051317819289067756,
+    "producer_p95_seconds": 0.1759442279999348,
+    "gpu_update_seconds": 0.968098927657382,
+    "producer_batches_per_second": 19.48640869494293,
+    "gpu_batches_per_second": 1.0329522855890487,
+    "producer_headroom": 18.86477136146774,
     "sustainable": true,
-    "buffer_seconds": 60.80909397330262,
+    "buffer_seconds": 60.022133514757684,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359076.50854694523,
+    "generation_bytes_per_second": 358002.935977222,
     "upload_bytes_per_second": 5915192.23919464,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.04945986086148,
+    "estimated_full_shards_per_hour": 76.81909617888923,
     "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": [

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

@@ -54,3 +54,4 @@
 {"step": 490400, "loss": 0.5646252036094666, "lr": 0.0003, "grad_norm": 0.16441540950719388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71778078272, "tokens_per_second": 145941.402206718, "data_wait_s": 1.1252999684074894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490500, "loss": 0.7770769596099854, "lr": 0.0003, "grad_norm": 0.40147290756149695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71792656749, "tokens_per_second": 152495.53326282685, "data_wait_s": 1.71370002135518e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490600, "loss": 0.5570622086524963, "lr": 0.0003, "grad_norm": 0.17094227700346842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71807382888, "tokens_per_second": 153244.69705833748, "data_wait_s": 1.0742000085883774e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 490700, "loss": 0.5813391208648682, "lr": 0.0003, "grad_norm": 0.1400664954006451, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71822127687, "tokens_per_second": 154452.1648729404, "data_wait_s": 1.6877998859854415e-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:c5e25fdc1425f1b3003dfde0603d66137ab7fb376cd7b3cc18e20078b431eb90
-size 119556
+oid sha256:342076a2bd4a39147d5dee474c1f8c918a54ae6a25f689e7e026e77ab3dbb4d5
+size 121674