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": 75275.90926623494,
-      "required_pages_per_second_upper_estimate": 0.006524725030149451
+      "required_token_positions_per_second": 75249.05969920107,
+      "required_pages_per_second_upper_estimate": 0.0065223977777285605
     },
     {
       "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": 9409.488658279368,
-      "required_pages_per_second_upper_estimate": 0.0037535857101800574
+      "required_token_positions_per_second": 9406.132462400134,
+      "required_pages_per_second_upper_estimate": 0.003752246873464231
     },
     {
       "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": 18818.977316558736,
-      "required_pages_per_second_upper_estimate": 0.004353439199497069
+      "required_token_positions_per_second": 18812.264924800267,
+      "required_pages_per_second_upper_estimate": 0.004351886405797807
     },
     {
       "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": 28228.465974838105,
-      "required_pages_per_second_upper_estimate": 0.001718965474552698
+      "required_token_positions_per_second": 28218.3973872004,
+      "required_pages_per_second_upper_estimate": 0.0017183523503913583
     },
     {
       "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": 28228.465974838105,
-      "required_pages_per_second_upper_estimate": 0.0021731054521381275
+      "required_token_positions_per_second": 28218.3973872004,
+      "required_pages_per_second_upper_estimate": 0.0021723303443901424
     },
     {
       "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": 28228.465974838105,
-      "required_pages_per_second_upper_estimate": 0.35050307280924425
+      "required_token_positions_per_second": 28218.3973872004,
+      "required_pages_per_second_upper_estimate": 0.35037805464817906
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.41369279896116495,
-    "producer_p95_seconds": 0.30607229299857863,
-    "gpu_update_seconds": 0.9794368572718223,
-    "producer_batches_per_second": 2.4172526147690427,
-    "gpu_batches_per_second": 1.0209948630945496,
-    "producer_headroom": 2.367546304241486,
+    "producer_batch_seconds": 0.4140352259844349,
+    "producer_p95_seconds": 0.33745574000204215,
+    "gpu_update_seconds": 0.9797863294866233,
+    "producer_batches_per_second": 2.4152534307252247,
+    "gpu_batches_per_second": 1.0206306925347366,
+    "producer_headroom": 2.366432293670242,
     "sustainable": true,
-    "buffer_seconds": 60.72508515085298,
+    "buffer_seconds": 60.746752428170645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 208100.13213754277,
-    "upload_bytes_per_second": 6030074.71480597,
+    "generation_bytes_per_second": 205198.1074427795,
+    "upload_bytes_per_second": 6119637.723563503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.65344403357231,
+    "estimated_full_shards_per_hour": 44.03073708975352,
     "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": 1.2982671780009696,
-      "bytes": 8839899,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.756701359001454,
       "bytes": 1071685,
@@ -562,6 +556,12 @@
       "bytes": 6010015,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3807983080005215,
+      "bytes": 15814135,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 18500, "loss": 0.7289327643811703, "lr": 0.0003, "grad_norm": 0.1548255705400022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2707145406, "tokens_per_second": 149209.71046541084, "data_wait_s": 9.69899701885879e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 18600, "loss": 0.6644454542547464, "lr": 0.0003, "grad_norm": 0.15401647829993756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2721860495, "tokens_per_second": 149481.10288003285, "data_wait_s": 1.3980999938212335e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 18700, "loss": 0.6925698667764664, "lr": 0.0003, "grad_norm": 0.15461326582955442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2736545256, "tokens_per_second": 149196.83346456004, "data_wait_s": 1.079399953596294e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 18800, "loss": 0.682854663580656, "lr": 0.0003, "grad_norm": 0.1324900247318256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2751283713, "tokens_per_second": 148798.14179879194, "data_wait_s": 1.4485001884168014e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2c13f9cd36c50cea54f80f93881e952652b5b5da57c577b9a3177d878f6db44
-size 392806
+oid sha256:33d3abd4f8b006d8dc16b279507fda090ea0a8ce3c6fc0855fdde826e5540fcb
+size 394924