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": 76537.07674522392,
-      "required_pages_per_second_upper_estimate": 0.005295012559719834
+      "required_token_positions_per_second": 76275.02857305737,
+      "required_pages_per_second_upper_estimate": 0.005276883459133821
     },
     {
       "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": 9567.13459315299,
-      "required_pages_per_second_upper_estimate": 0.0040055594696646
+      "required_token_positions_per_second": 9534.37857163217,
+      "required_pages_per_second_upper_estimate": 0.003991845207477346
     },
     {
       "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": 19134.26918630598,
-      "required_pages_per_second_upper_estimate": 0.003985163666895902
+      "required_token_positions_per_second": 19068.75714326434,
+      "required_pages_per_second_upper_estimate": 0.003971519235999085
     },
     {
       "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": 28701.40377945897,
-      "required_pages_per_second_upper_estimate": 0.0017206053320725154
+      "required_token_positions_per_second": 28603.13571489651,
+      "required_pages_per_second_upper_estimate": 0.0017147143116486424
     },
     {
       "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": 28701.40377945897,
-      "required_pages_per_second_upper_estimate": 0.0022314078997196004
+      "required_token_positions_per_second": 28603.13571489651,
+      "required_pages_per_second_upper_estimate": 0.002223767989935405
     },
     {
       "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": 28701.40377945897,
-      "required_pages_per_second_upper_estimate": 0.3681507905165271
+      "required_token_positions_per_second": 28603.13571489651,
+      "required_pages_per_second_upper_estimate": 0.366890313296347
     }
   ],
   "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": 0.17516045093745447,
-    "producer_p95_seconds": 0.2660683389985934,
-    "gpu_update_seconds": 0.9632978307418931,
-    "producer_batches_per_second": 5.7090512992403495,
-    "gpu_batches_per_second": 1.0381005417917741,
-    "producer_headroom": 5.499516732152416,
+    "producer_batch_seconds": 0.09226781845310938,
+    "producer_p95_seconds": 0.27754543499759166,
+    "gpu_update_seconds": 0.9666073075198158,
+    "producer_batches_per_second": 10.83801499553391,
+    "gpu_batches_per_second": 1.0345462859843935,
+    "producer_headroom": 10.476104493692421,
     "sustainable": true,
-    "buffer_seconds": 60.687763336739266,
+    "buffer_seconds": 60.8962603737484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 624390.3036111611,
+    "generation_bytes_per_second": 636033.0404543991,
     "upload_bytes_per_second": 4913846.067093352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.9898120462948,
+    "estimated_full_shards_per_hour": 68.238942195053,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 906400, "loss": 0.5510749220848083, "lr": 3e-05, "grad_norm": 0.33704090464142944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132671214845, "tokens_per_second": 151571.010057406, "data_wait_s": 1.7296999430982396e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906500, "loss": 0.5194767713546753, "lr": 3e-05, "grad_norm": 0.319990316796002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132685900827, "tokens_per_second": 152826.49416012273, "data_wait_s": 1.9835002603940666e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906600, "loss": 0.527674674987793, "lr": 3e-05, "grad_norm": 0.34377388706849443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132700371678, "tokens_per_second": 150210.87395500537, "data_wait_s": 1.0879000910790637e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 906700, "loss": 0.5318596363067627, "lr": 3e-05, "grad_norm": 0.3319238812255569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132714852166, "tokens_per_second": 150475.67974385415, "data_wait_s": 1.4097000530455261e-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 sha256:7a6f62b10ff5058d8c6f7c69b9f486d42da4b4b68e3e4a963834bf2364412dfd
-size 460785
+oid sha256:4ca7c7ba62bac8da894ac9abd2ecc0251f1b5868213e0c4abc9bc1f62c112e0c
+size 462903