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": 75594.44660321492,
-      "required_pages_per_second_upper_estimate": 0.006552335039193376
+      "required_token_positions_per_second": 75969.27408228614,
+      "required_pages_per_second_upper_estimate": 0.006584824135087709
     },
     {
       "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": 9449.305825401865,
-      "required_pages_per_second_upper_estimate": 0.0037694693734649214
+      "required_token_positions_per_second": 9496.159260285767,
+      "required_pages_per_second_upper_estimate": 0.003788159909161388
     },
     {
       "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": 18898.61165080373,
-      "required_pages_per_second_upper_estimate": 0.004371861201208233
+      "required_token_positions_per_second": 18992.318520571534,
+      "required_pages_per_second_upper_estimate": 0.0043935386363444335
     },
     {
       "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": 28347.9174762056,
-      "required_pages_per_second_upper_estimate": 0.0017262394442723857
+      "required_token_positions_per_second": 28488.477780857298,
+      "required_pages_per_second_upper_estimate": 0.0017347988293627498
     },
     {
       "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": 28347.9174762056,
-      "required_pages_per_second_upper_estimate": 0.002182301159376316
+      "required_token_positions_per_second": 28488.477780857298,
+      "required_pages_per_second_upper_estimate": 0.0021931218807242307
     },
     {
       "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": 28347.9174762056,
-      "required_pages_per_second_upper_estimate": 0.35198626067777045
+      "required_token_positions_per_second": 28488.477780857298,
+      "required_pages_per_second_upper_estimate": 0.35373154923646644
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11876503291426843,
-    "producer_p95_seconds": 0.39958601900434587,
-    "gpu_update_seconds": 0.9753097391794975,
-    "producer_batches_per_second": 8.419986720517803,
-    "gpu_batches_per_second": 1.0253153022354455,
-    "producer_headroom": 8.21209505228305,
+    "producer_batch_seconds": 0.13745842416454934,
+    "producer_p95_seconds": 0.421831995001412,
+    "gpu_update_seconds": 0.9704976240807764,
+    "producer_batches_per_second": 7.274926990308834,
+    "gpu_batches_per_second": 1.0303992252914242,
+    "producer_headroom": 7.060299359455837,
     "sustainable": true,
-    "buffer_seconds": 60.469203829128844,
+    "buffer_seconds": 60.170852693008136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382989.49974669114,
-    "upload_bytes_per_second": 5686576.56180762,
+    "generation_bytes_per_second": 382384.3625276046,
+    "upload_bytes_per_second": 5692026.758920163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.1806311063819,
+    "estimated_full_shards_per_hour": 82.05078274603943,
     "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": 1.4038716970026144,
-      "bytes": 7663039,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9987416639996809,
       "bytes": 1128633,
@@ -562,6 +556,12 @@
       "bytes": 5169545,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.040382787999988,
+      "bytes": 5992429,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -340,3 +340,4 @@
 {"step": 314200, "loss": 0.5247049517929554, "lr": 0.0003, "grad_norm": 0.16029382198406003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45990174774, "tokens_per_second": 148771.5261618356, "data_wait_s": 1.0413998097646981e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314300, "loss": 0.5914095975458622, "lr": 0.0003, "grad_norm": 0.1586909353515651, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46004678255, "tokens_per_second": 148714.66425481616, "data_wait_s": 1.3215998478699476e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314400, "loss": 0.572323065251112, "lr": 0.0003, "grad_norm": 0.15096272257554857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46019294449, "tokens_per_second": 149564.04520477745, "data_wait_s": 1.708900526864454e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 314500, "loss": 0.5538896843791008, "lr": 0.0003, "grad_norm": 0.14577745322222035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46033954052, "tokens_per_second": 149969.83623488643, "data_wait_s": 9.750001481734216e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:833ae04237735897cf8159d702397c21070f599a24d3b5c67ab5d4a22c20a555
-size 729219
+oid sha256:f11a7db5b34c9debae4c2782041a38165ce88dc2ece62db69ff6cc24d2e770e7
+size 731337