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": 76372.10071417736,
-      "required_pages_per_second_upper_estimate": 0.005283599135094901
+      "required_token_positions_per_second": 76674.44585513709,
+      "required_pages_per_second_upper_estimate": 0.005304516073483876
     },
     {
       "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": 9546.51258927217,
-      "required_pages_per_second_upper_estimate": 0.003996925467276111
+      "required_token_positions_per_second": 9584.305731892136,
+      "required_pages_per_second_upper_estimate": 0.004012748666880529
     },
     {
       "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": 19093.02517854434,
-      "required_pages_per_second_upper_estimate": 0.00397657362775678
+      "required_token_positions_per_second": 19168.611463784273,
+      "required_pages_per_second_upper_estimate": 0.0039923162576278705
     },
     {
       "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": 28639.53776781651,
-      "required_pages_per_second_upper_estimate": 0.00171689655913848
+      "required_token_positions_per_second": 28752.917195676408,
+      "required_pages_per_second_upper_estimate": 0.0017236934827183183
     },
     {
       "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": 28639.53776781651,
-      "required_pages_per_second_upper_estimate": 0.002226598089434226
+      "required_token_positions_per_second": 28752.917195676408,
+      "required_pages_per_second_upper_estimate": 0.0022354128412469345
     },
     {
       "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": 28639.53776781651,
-      "required_pages_per_second_upper_estimate": 0.36735723974572554
+      "required_token_positions_per_second": 28752.917195676408,
+      "required_pages_per_second_upper_estimate": 0.3688115493089674
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09083926948451904,
-    "producer_p95_seconds": 0.258365114001208,
-    "gpu_update_seconds": 0.9653787091169208,
-    "producer_batches_per_second": 11.008454886027254,
-    "gpu_batches_per_second": 1.0358629111623447,
-    "producer_headroom": 10.62732796724485,
+    "producer_batch_seconds": 0.07151475482820047,
+    "producer_p95_seconds": 0.24986580699987826,
+    "gpu_update_seconds": 0.9615719967418626,
+    "producer_batches_per_second": 13.98312841038601,
+    "gpu_batches_per_second": 1.0399637295889905,
+    "producer_headroom": 13.445784706272741,
     "sustainable": true,
-    "buffer_seconds": 60.81885867436601,
+    "buffer_seconds": 60.579035794737344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321985.95348061604,
+    "generation_bytes_per_second": 322012.32755086216,
     "upload_bytes_per_second": 4864276.924059098,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.09069016755925,
+    "estimated_full_shards_per_hour": 69.09634942907714,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 785400, "loss": 0.5478571653366089, "lr": 3e-05, "grad_norm": 0.2627609841850018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114960231236, "tokens_per_second": 151862.56571565307, "data_wait_s": 1.1579999409150332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 785500, "loss": 0.5633331537246704, "lr": 3e-05, "grad_norm": 0.2577616388919846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114974904358, "tokens_per_second": 152468.21620589224, "data_wait_s": 1.5104997146409005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 785600, "loss": 0.5228918790817261, "lr": 3e-05, "grad_norm": 0.2482094230685791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114989536499, "tokens_per_second": 151982.92673010292, "data_wait_s": 7.004000508459285e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 785700, "loss": 0.532866358757019, "lr": 3e-05, "grad_norm": 0.2698860448119994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115004283662, "tokens_per_second": 153080.66404407044, "data_wait_s": 1.0274998203385621e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe8d203912750a32d7c617ece6f5f54f0cccf136cbf17c4a6c0f55bc65156901
-size 546288
+oid sha256:32144ee3b8508c82c8ff029afbbca71620c3a86c5d578339000651286983ad26
+size 548406