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": 76144.38235481431,
-      "required_pages_per_second_upper_estimate": 0.006600002076342592
+      "required_token_positions_per_second": 76292.21369091657,
+      "required_pages_per_second_upper_estimate": 0.006612815721880839
     },
     {
       "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": 9518.047794351789,
-      "required_pages_per_second_upper_estimate": 0.003796891572663072
+      "required_token_positions_per_second": 9536.526711364571,
+      "required_pages_per_second_upper_estimate": 0.0038042630889439013
     },
     {
       "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": 19036.095588703578,
-      "required_pages_per_second_upper_estimate": 0.004403665690606696
+      "required_token_positions_per_second": 19073.053422729143,
+      "required_pages_per_second_upper_estimate": 0.004412215235072847
     },
     {
       "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": 28554.143383055365,
-      "required_pages_per_second_upper_estimate": 0.0017387975200158232
+      "required_token_positions_per_second": 28609.58013409371,
+      "required_pages_per_second_upper_estimate": 0.001742173327299904
     },
     {
       "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": 28554.143383055365,
-      "required_pages_per_second_upper_estimate": 0.002198177000555459
+      "required_token_positions_per_second": 28609.58013409371,
+      "required_pages_per_second_upper_estimate": 0.002202444675109139
     },
     {
       "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": 28554.143383055365,
-      "required_pages_per_second_upper_estimate": 0.3545468962471332
+      "required_token_positions_per_second": 28609.58013409371,
+      "required_pages_per_second_upper_estimate": 0.3552352351601588
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.42046161182032904,
-    "producer_p95_seconds": 0.21737056399979338,
-    "gpu_update_seconds": 0.9682657829758924,
-    "producer_batches_per_second": 2.3783384068539375,
-    "gpu_batches_per_second": 1.0327742832412965,
-    "producer_headroom": 2.3028636996940643,
+    "producer_batch_seconds": 0.07766030386714817,
+    "producer_p95_seconds": 0.18840971600002376,
+    "gpu_update_seconds": 0.9663895754643457,
+    "producer_batches_per_second": 12.876591388448322,
+    "gpu_batches_per_second": 1.034779374062996,
+    "producer_headroom": 12.443803685310423,
     "sustainable": true,
-    "buffer_seconds": 60.03247854450533,
+    "buffer_seconds": 60.882543254253775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276412.1282724496,
-    "upload_bytes_per_second": 5521329.33686206,
+    "generation_bytes_per_second": 286922.3171266021,
+    "upload_bytes_per_second": 5537343.747959721,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.31160818224064,
+    "estimated_full_shards_per_hour": 61.566850045667145,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 10399159,
       "pages": 23,
       "shards": 3
+    },
+    {
+      "seconds": 0.9434353479991842,
+      "bytes": 5975783,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -47,3 +47,4 @@
 {"step": 284900, "loss": 0.5730718392878771, "lr": 0.0003, "grad_norm": 0.16240993122157482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41701454375, "tokens_per_second": 149735.90690109433, "data_wait_s": 1.4601999282604083e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285000, "loss": 0.5920775551348925, "lr": 0.0003, "grad_norm": 0.15936638618640356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41715928091, "tokens_per_second": 146928.87176656383, "data_wait_s": 2.6560001060715877e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 285100, "loss": 0.5674477610737085, "lr": 0.0003, "grad_norm": 0.15156043304043362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41730571018, "tokens_per_second": 130993.04942865677, "data_wait_s": 1.2206999599584378e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 285200, "loss": 0.659591244533658, "lr": 0.0003, "grad_norm": 0.1874568876131716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41745218309, "tokens_per_second": 149745.86236816508, "data_wait_s": 1.4650999219156802e-05, "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:ea8db65f33c08e0c78e29201dd603442a6c77431fa9b6f5113ebb369a230b5b7
-size 104730
+oid sha256:71f8e3469f78b96608117d7fd6ced8fd32941443565c7b3e6d1d3532014ad8ed
+size 106848