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": 76552.48050270948,
-      "required_pages_per_second_upper_estimate": 0.005296078227404326
+      "required_token_positions_per_second": 76335.24010499836,
+      "required_pages_per_second_upper_estimate": 0.0052810490326235015
     },
     {
       "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": 9569.060062838686,
-      "required_pages_per_second_upper_estimate": 0.004006365623613622
+      "required_token_positions_per_second": 9541.905013124795,
+      "required_pages_per_second_upper_estimate": 0.003994996371360336
     },
     {
       "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": 19138.12012567737,
-      "required_pages_per_second_upper_estimate": 0.003985965716010863
+      "required_token_positions_per_second": 19083.81002624959,
+      "required_pages_per_second_upper_estimate": 0.003974654354553692
     },
     {
       "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": 28707.180188516057,
-      "required_pages_per_second_upper_estimate": 0.0017209516189754216
+      "required_token_positions_per_second": 28625.715039374383,
+      "required_pages_per_second_upper_estimate": 0.0017160679076744585
     },
     {
       "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": 28707.180188516057,
-      "required_pages_per_second_upper_estimate": 0.0022318569901160507
+      "required_token_positions_per_second": 28625.715039374383,
+      "required_pages_per_second_upper_estimate": 0.0022255234330976073
     },
     {
       "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": 28707.180188516057,
-      "required_pages_per_second_upper_estimate": 0.3682248840896866
+      "required_token_positions_per_second": 28625.715039374383,
+      "required_pages_per_second_upper_estimate": 0.3671799366269594
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2645496175702533,
+    "producer_batch_seconds": 0.3096811289844368,
     "producer_p95_seconds": 0.29012310200050706,
-    "gpu_update_seconds": 0.9631039976214812,
-    "producer_batches_per_second": 3.7800092443318003,
-    "gpu_batches_per_second": 1.0383094686239893,
-    "producer_headroom": 3.6405420142621114,
+    "gpu_update_seconds": 0.9658448692712288,
+    "producer_batches_per_second": 3.229127984902999,
+    "gpu_batches_per_second": 1.035362957153298,
+    "producer_headroom": 3.1188366964387035,
     "sustainable": true,
-    "buffer_seconds": 60.67555185015332,
+    "buffer_seconds": 60.848226764087414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516008.185264992,
-    "upload_bytes_per_second": 5104712.42265679,
+    "generation_bytes_per_second": 481796.7567845915,
+    "upload_bytes_per_second": 5101057.749060589,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.72334450208969,
+    "estimated_full_shards_per_hour": 103.38236835149105,
     "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": 1.3309355630008213,
-      "bytes": 11158362,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9890253090015904,
       "bytes": 2755782,
@@ -562,6 +556,12 @@
       "bytes": 887331,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3929757290025009,
+      "bytes": 11170736,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 753400, "loss": 0.5324732065200806, "lr": 5e-05, "grad_norm": 0.23146167856273578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110276564166, "tokens_per_second": 150409.29126215793, "data_wait_s": 9.796996891964227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 753500, "loss": 0.5384315252304077, "lr": 5e-05, "grad_norm": 0.23645750828588763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110291281084, "tokens_per_second": 151705.62844042058, "data_wait_s": 1.3329998182598501e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 753600, "loss": 0.4969802498817444, "lr": 5e-05, "grad_norm": 0.2267397935405334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110305946145, "tokens_per_second": 151113.381995992, "data_wait_s": 1.5074998373165727e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 753700, "loss": 0.5267372131347656, "lr": 5e-05, "grad_norm": 0.24205623519582248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110320734202, "tokens_per_second": 152452.91525136866, "data_wait_s": 1.2331001926213503e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb660ee759424e8bb61779ce0158b61fbb2dce7b96f86d93f4b5b1b5b8c58dd4
-size 716511
+oid sha256:888ecee7541c0c6e2d24c60b86b5234491be7bafa3710933ac40cd115a77861f
+size 718629