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": 75659.00007784228,
-      "required_pages_per_second_upper_estimate": 0.00655793037076479
+      "required_token_positions_per_second": 75757.21789836236,
+      "required_pages_per_second_upper_estimate": 0.00656644364251668
     },
     {
       "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": 9457.375009730285,
-      "required_pages_per_second_upper_estimate": 0.0037726882917385853
+      "required_token_positions_per_second": 9469.652237295295,
+      "required_pages_per_second_upper_estimate": 0.0037775858613751773
     },
     {
       "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": 18914.75001946057,
-      "required_pages_per_second_upper_estimate": 0.004375594528771405
+      "required_token_positions_per_second": 18939.30447459059,
+      "required_pages_per_second_upper_estimate": 0.004381274769822085
     },
     {
       "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": 28372.125029190855,
-      "required_pages_per_second_upper_estimate": 0.0017277135572419727
+      "required_token_positions_per_second": 28408.956711885887,
+      "required_pages_per_second_upper_estimate": 0.0017299564134771956
     },
     {
       "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": 28372.125029190855,
-      "required_pages_per_second_upper_estimate": 0.0021841647238159166
+      "required_token_positions_per_second": 28408.956711885887,
+      "required_pages_per_second_upper_estimate": 0.0021870001287064037
     },
     {
       "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": 28372.125029190855,
-      "required_pages_per_second_upper_estimate": 0.3522868374683792
+      "required_token_positions_per_second": 28408.956711885887,
+      "required_pages_per_second_upper_estimate": 0.3527441636997391
     }
   ],
   "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": 0.10635448780467982,
-    "producer_p95_seconds": 0.240300378000029,
-    "gpu_update_seconds": 0.9744775892378229,
-    "producer_batches_per_second": 9.402518131971089,
-    "gpu_batches_per_second": 1.0261908647710813,
-    "producer_headroom": 9.162543202008104,
+    "producer_batch_seconds": 0.07958720800785102,
+    "producer_p95_seconds": 0.2192232369998237,
+    "gpu_update_seconds": 0.9732141972124054,
+    "producer_batches_per_second": 12.564833282018805,
+    "gpu_batches_per_second": 1.0275230292204096,
+    "producer_headroom": 12.228274135667645,
     "sustainable": true,
-    "buffer_seconds": 60.41761053274502,
+    "buffer_seconds": 60.339280227169134,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382707.26627937413,
-    "upload_bytes_per_second": 4732548.3317361465,
+    "generation_bytes_per_second": 484660.31643388374,
+    "upload_bytes_per_second": 4672730.834553713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.12007037435454,
+    "estimated_full_shards_per_hour": 103.99682159197219,
     "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": 0.8288988310005152,
-      "bytes": 5801101,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2185628970000835,
       "bytes": 7809668,
@@ -562,6 +556,12 @@
       "bytes": 7646685,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9418184709993511,
+      "bytes": 1092329,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 250000, "loss": 0.5805718433111906, "lr": 0.0003, "grad_norm": 0.15552392142350027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36593230586, "tokens_per_second": 150453.01984795262, "data_wait_s": 1.3057999240118079e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250100, "loss": 0.568646140396595, "lr": 0.0003, "grad_norm": 0.15801772858695023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36607938466, "tokens_per_second": 136372.44597009715, "data_wait_s": 1.2935999620822258e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 250200, "loss": 0.6084444671869278, "lr": 0.0003, "grad_norm": 0.1512300028949969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36622541903, "tokens_per_second": 149158.94468358142, "data_wait_s": 1.3777998901787214e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 250300, "loss": 0.5982154533267021, "lr": 0.0003, "grad_norm": 0.14984847969284623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36637338504, "tokens_per_second": 151302.56202830066, "data_wait_s": 1.4895998901920393e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e87d4d07e0217a68716cb3074f7ef0b211385f793ec045520fd0acc5f18f6664
-size 279972
+oid sha256:9ae4f083465a29dd4beb5e3c161bad1710ffa61553a48c794ce921d642847404
+size 282090