Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77140.35225317664,
-      "required_pages_per_second_upper_estimate": 0.005336748559152035
+      "required_token_positions_per_second": 76846.38526672727,
+      "required_pages_per_second_upper_estimate": 0.005316411240932084
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9642.54403164708,
-      "required_pages_per_second_upper_estimate": 0.004037131826833931
+      "required_token_positions_per_second": 9605.79815834091,
+      "required_pages_per_second_upper_estimate": 0.00402174709702173
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19285.08806329416,
-      "required_pages_per_second_upper_estimate": 0.004016575261611288
+      "required_token_positions_per_second": 19211.59631668182,
+      "required_pages_per_second_upper_estimate": 0.004001268868899627
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28927.63209494124,
-      "required_pages_per_second_upper_estimate": 0.0017341673741550407
+      "required_token_positions_per_second": 28817.394475022724,
+      "required_pages_per_second_upper_estimate": 0.001727558797164025
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28927.63209494124,
-      "required_pages_per_second_upper_estimate": 0.0022489961561751497
+      "required_token_positions_per_second": 28817.394475022724,
+      "required_pages_per_second_upper_estimate": 0.0022404256661105308
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28927.63209494124,
-      "required_pages_per_second_upper_estimate": 0.37105260444249355
+      "required_token_positions_per_second": 28817.394475022724,
+      "required_pages_per_second_upper_estimate": 0.3696385946181132
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11941365863285114,
-    "producer_p95_seconds": 0.2406009599999379,
-    "gpu_update_seconds": 0.9557643677594677,
-    "producer_batches_per_second": 8.374251416871807,
-    "gpu_batches_per_second": 1.046282989545039,
-    "producer_headroom": 8.00381111090531,
+    "producer_batch_seconds": 0.14265774477337345,
+    "producer_p95_seconds": 0.19716906400026346,
+    "gpu_update_seconds": 0.9594205341486965,
+    "producer_batches_per_second": 7.009784162707767,
+    "gpu_batches_per_second": 1.0422958071116437,
+    "producer_headroom": 6.72533086565216,
     "sustainable": true,
-    "buffer_seconds": 60.21315516884646,
+    "buffer_seconds": 60.44349365136788,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288977.5374915778,
-    "upload_bytes_per_second": 5096524.265813544,
+    "generation_bytes_per_second": 308945.9180816672,
+    "upload_bytes_per_second": 5065123.461648618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.007852492909436,
+    "estimated_full_shards_per_hour": 66.29260212743294,
     "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": 0.9894983370004411,
-      "bytes": 7420139,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.4091035430001284,
       "bytes": 10556108,
@@ -562,6 +556,12 @@
       "bytes": 2360936,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2458887120010331,
+      "bytes": 6052831,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -124,3 +124,4 @@
 {"step": 452600, "loss": 0.6468116641044617, "lr": 0.0003, "grad_norm": 0.1909410882565782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66246885961, "tokens_per_second": 152780.04444875772, "data_wait_s": 1.416400118614547e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452700, "loss": 0.6137679815292358, "lr": 0.0003, "grad_norm": 0.16246855474517732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66261504772, "tokens_per_second": 151020.6673284527, "data_wait_s": 1.5814999642316252e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 452800, "loss": 0.5833898186683655, "lr": 0.0003, "grad_norm": 0.1638339779408624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66275983640, "tokens_per_second": 151077.60845432733, "data_wait_s": 1.3533999663195573e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 452900, "loss": 0.5924435257911682, "lr": 0.0003, "grad_norm": 0.19952839026130523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66290400713, "tokens_per_second": 150704.25808922146, "data_wait_s": 1.7930999092641287e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dada0341ba096269fccd6c60d0caf1c593d390355ff0f981454cb37a70a31212
-size 268599
+oid sha256:310755e8e4c7ddf37ad0ba80383f324420ac8eb46c9614d52c06d7e311458e71
+size 270717