Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76476.08831885221,
-      "required_pages_per_second_upper_estimate": 0.006628753508604317
+      "required_token_positions_per_second": 76435.90399737709,
+      "required_pages_per_second_upper_estimate": 0.006625270433465085
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9559.511039856527,
-      "required_pages_per_second_upper_estimate": 0.003813431881225677
+      "required_token_positions_per_second": 9554.487999672136,
+      "required_pages_per_second_upper_estimate": 0.00381142811539498
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19119.022079713053,
-      "required_pages_per_second_upper_estimate": 0.004422849274845343
+      "required_token_positions_per_second": 19108.97599934427,
+      "required_pages_per_second_upper_estimate": 0.004420525290957001
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28678.53311956958,
-      "required_pages_per_second_upper_estimate": 0.0017463722023470286
+      "required_token_positions_per_second": 28663.463999016407,
+      "required_pages_per_second_upper_estimate": 0.001745454571966906
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28678.53311956958,
-      "required_pages_per_second_upper_estimate": 0.002207752866805152
+      "required_token_positions_per_second": 28663.463999016407,
+      "required_pages_per_second_upper_estimate": 0.002206592804190973
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28678.53311956958,
-      "required_pages_per_second_upper_estimate": 0.35609140046897175
+      "required_token_positions_per_second": 28663.463999016407,
+      "required_pages_per_second_upper_estimate": 0.3559042924248036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11805967920315652,
-    "producer_p95_seconds": 0.3413104540013592,
-    "gpu_update_seconds": 0.9640660449656553,
-    "producer_batches_per_second": 8.470292370346058,
-    "gpu_batches_per_second": 1.0372733333177655,
-    "producer_headroom": 8.165921265182291,
+    "producer_batch_seconds": 0.13402592072662856,
+    "producer_p95_seconds": 0.3446127639981569,
+    "gpu_update_seconds": 0.9645728792915171,
+    "producer_batches_per_second": 7.461243277258963,
+    "gpu_batches_per_second": 1.0367282985755355,
+    "producer_headroom": 7.196912911040154,
     "sustainable": true,
-    "buffer_seconds": 60.73616083283628,
+    "buffer_seconds": 60.76809139536558,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372935.095843462,
-    "upload_bytes_per_second": 4272189.235532882,
+    "generation_bytes_per_second": 386939.605730386,
+    "upload_bytes_per_second": 4314448.409949572,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.0231900832929,
+    "estimated_full_shards_per_hour": 83.02823189672169,
     "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": 1.1510198380019574,
-      "bytes": 2279083,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 19.352378830000816,
       "bytes": 15868354,
@@ -562,6 +556,12 @@
       "bytes": 235889,
       "pages": 8,
       "shards": 1
+    },
+    {
+      "seconds": 1.3448229619971244,
+      "bytes": 7303500,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -350,3 +350,4 @@
 {"step": 190000, "loss": 0.5794018357992172, "lr": 0.0003, "grad_norm": 0.143421867137207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27808256418, "tokens_per_second": 150552.32028538815, "data_wait_s": 7.763999747112393e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190100, "loss": 0.5956941954791546, "lr": 0.0003, "grad_norm": 0.15725801850204055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27822802377, "tokens_per_second": 120622.25517734191, "data_wait_s": 8.716000593267381e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190200, "loss": 0.6160158794373274, "lr": 0.0003, "grad_norm": 0.1624706316048518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27837334859, "tokens_per_second": 149570.92182152078, "data_wait_s": 1.1028001608792692e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 190300, "loss": 0.6052691824734211, "lr": 0.0003, "grad_norm": 0.1512022836333551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27851959384, "tokens_per_second": 149854.4557606955, "data_wait_s": 1.0021998605225235e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4d9b4ea384fe59ca973799ffd6e935e70ccd4289fcaba96bf25ed6a3a1c5572
-size 751182
+oid sha256:2638b3a000b2187ff509e6173c577dcd49959b3ef01363826ac222ca30074131
+size 753300