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": 74877.65248182749,
-      "required_pages_per_second_upper_estimate": 0.006490205141449605
+      "required_token_positions_per_second": 74058.533816883,
+      "required_pages_per_second_upper_estimate": 0.006419206011609494
     },
     {
       "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": 9359.706560228437,
-      "required_pages_per_second_upper_estimate": 0.0037337268869588464
+      "required_token_positions_per_second": 9257.316727110376,
+      "required_pages_per_second_upper_estimate": 0.0036928820516636254
     },
     {
       "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": 18719.413120456873,
-      "required_pages_per_second_upper_estimate": 0.004330406774998906
+      "required_token_positions_per_second": 18514.63345422075,
+      "required_pages_per_second_upper_estimate": 0.004283034603214214
     },
     {
       "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": 28079.119680685308,
-      "required_pages_per_second_upper_estimate": 0.0017098710687982414
+      "required_token_positions_per_second": 27771.950181331125,
+      "required_pages_per_second_upper_estimate": 0.0016911660578813842
     },
     {
       "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": 28079.119680685308,
-      "required_pages_per_second_upper_estimate": 0.002161608360997243
+      "required_token_positions_per_second": 27771.950181331125,
+      "required_pages_per_second_upper_estimate": 0.002137961602637362
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28079.119680685308,
-      "required_pages_per_second_upper_estimate": 0.34864869166575996
+      "required_token_positions_per_second": 27771.950181331125,
+      "required_pages_per_second_upper_estimate": 0.3448346745139641
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.097406510632851,
-    "producer_p95_seconds": 0.31932546800089767,
-    "gpu_update_seconds": 0.9846462536721954,
-    "producer_batches_per_second": 10.266254211376538,
-    "gpu_batches_per_second": 1.0155931597470091,
-    "producer_headroom": 10.108628748478306,
+    "producer_batch_seconds": 0.4830811388123948,
+    "producer_p95_seconds": 0.33351783100079047,
+    "gpu_update_seconds": 0.9955368571338412,
+    "producer_batches_per_second": 2.0700456293085607,
+    "gpu_batches_per_second": 1.0044831518131918,
+    "producer_headroom": 2.0608067199254894,
     "sustainable": true,
-    "buffer_seconds": 60.06342147400392,
+    "buffer_seconds": 60.72774828516432,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467036.96441197745,
+    "generation_bytes_per_second": 417008.00709706434,
     "upload_bytes_per_second": 4102907.7703114543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.21526049870961,
+    "estimated_full_shards_per_hour": 89.48021087345073,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 104900, "loss": 0.5931651797145605, "lr": 0.0003, "grad_norm": 0.12879474546570285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15350920625, "tokens_per_second": 148624.37984174673, "data_wait_s": 1.4916997315594926e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105000, "loss": 0.6216882299631834, "lr": 0.0003, "grad_norm": 0.13170376331896053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15365640815, "tokens_per_second": 148940.21746605175, "data_wait_s": 1.3888999092159793e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105100, "loss": 0.6351448111236095, "lr": 0.0003, "grad_norm": 0.1474901384848249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15380211498, "tokens_per_second": 124904.54535929876, "data_wait_s": 1.1435997294029221e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 105200, "loss": 0.6166014447808266, "lr": 0.0003, "grad_norm": 0.1407138276989707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15394739203, "tokens_per_second": 147867.80495907145, "data_wait_s": 1.6666999727021903e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93d4f3733ed4f94113132dd6c7e7b55127961923fadf488b2769fe3a78d423ad
-size 429015
+oid sha256:3e351ea1ebdf5eaf2c58376a3e8a5932fd63733c55d02839feb689af3babfbd6
+size 431133