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": 75810.75598818513,
-      "required_pages_per_second_upper_estimate": 0.005244764004451546
+      "required_token_positions_per_second": 76181.67732078354,
+      "required_pages_per_second_upper_estimate": 0.00527042520289677
     },
     {
       "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": 9476.344498523142,
-      "required_pages_per_second_upper_estimate": 0.0039675475529558505
+      "required_token_positions_per_second": 9522.709665097942,
+      "required_pages_per_second_upper_estimate": 0.003986959679985942
     },
     {
       "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": 18952.688997046283,
-      "required_pages_per_second_upper_estimate": 0.0039473453020647175
+      "required_token_positions_per_second": 19045.419330195884,
+      "required_pages_per_second_upper_estimate": 0.003966658584996443
     },
     {
       "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": 28429.033495569423,
-      "required_pages_per_second_upper_estimate": 0.0017042771494386799
+      "required_token_positions_per_second": 28568.128995293828,
+      "required_pages_per_second_upper_estimate": 0.0017126157122606274
     },
     {
       "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": 28429.033495569423,
-      "required_pages_per_second_upper_estimate": 0.0022102323081774553
+      "required_token_positions_per_second": 28568.128995293828,
+      "required_pages_per_second_upper_estimate": 0.002221046371464583
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28429.033495569423,
-      "required_pages_per_second_upper_estimate": 0.3646571169632178
+      "required_token_positions_per_second": 28568.128995293828,
+      "required_pages_per_second_upper_estimate": 0.3664412846845708
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12072705212511892,
-    "producer_p95_seconds": 0.2712036260018067,
-    "gpu_update_seconds": 0.972526906491874,
-    "producer_batches_per_second": 8.283147665724675,
-    "gpu_batches_per_second": 1.0282491860376672,
-    "producer_headroom": 8.055583975362605,
+    "producer_batch_seconds": 0.07346481235941837,
+    "producer_p95_seconds": 0.2612655189986981,
+    "gpu_update_seconds": 0.9677917655914602,
+    "producer_batches_per_second": 13.61195881243951,
+    "gpu_batches_per_second": 1.0332801285913567,
+    "producer_headroom": 13.17354165224907,
     "sustainable": true,
-    "buffer_seconds": 60.296668202496186,
+    "buffer_seconds": 60.00308946667053,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379882.6176367886,
-    "upload_bytes_per_second": 4881374.903585045,
+    "generation_bytes_per_second": 420976.49528059666,
+    "upload_bytes_per_second": 4853238.968750135,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.51396653011076,
+    "estimated_full_shards_per_hour": 90.33175605595994,
     "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": 1.339241851999759,
-      "bytes": 10779600,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 4.318171066001014,
       "bytes": 1094935,
@@ -562,6 +556,12 @@
       "bytes": 1216935,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.176057447999483,
+      "bytes": 7472528,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 783000, "loss": 0.5296126008033752, "lr": 3e-05, "grad_norm": 0.26037759137340166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114609159455, "tokens_per_second": 152681.0293511491, "data_wait_s": 1.2995998986298218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783100, "loss": 0.5236709117889404, "lr": 3e-05, "grad_norm": 0.25507218812778376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114623872966, "tokens_per_second": 152265.31835006358, "data_wait_s": 1.296800110139884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783200, "loss": 0.5158957839012146, "lr": 3e-05, "grad_norm": 0.24981190451559307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114638666763, "tokens_per_second": 153343.64682076938, "data_wait_s": 1.4785997336730361e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 783300, "loss": 0.49220219254493713, "lr": 3e-05, "grad_norm": 0.26907296387154755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114653339897, "tokens_per_second": 152946.0809744229, "data_wait_s": 1.1788000847445801e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da234903c7b62d858a9c62b53bdc8e83e2d2e89f3dae78fa56050de422e79745
-size 494673
+oid sha256:0b9ff589a7de459f4d2dc273ac7829f252bea790a83ed72651e1442f63207065
+size 496791