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": 76517.60648802576,
-      "required_pages_per_second_upper_estimate": 0.005293665562149644
+      "required_token_positions_per_second": 76313.75839753207,
+      "required_pages_per_second_upper_estimate": 0.005279562878256555
     },
     {
       "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": 9564.70081100322,
-      "required_pages_per_second_upper_estimate": 0.004004540495901643
+      "required_token_positions_per_second": 9539.21979969151,
+      "required_pages_per_second_upper_estimate": 0.003993872128569453
     },
     {
       "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": 19129.40162200644,
-      "required_pages_per_second_upper_estimate": 0.003984149881618611
+      "required_token_positions_per_second": 19078.43959938302,
+      "required_pages_per_second_upper_estimate": 0.003973535836265066
     },
     {
       "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": 28694.102433009662,
-      "required_pages_per_second_upper_estimate": 0.0017201676275013897
+      "required_token_positions_per_second": 28617.659399074528,
+      "required_pages_per_second_upper_estimate": 0.0017155849843387327
     },
     {
       "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": 28694.102433009662,
-      "required_pages_per_second_upper_estimate": 0.002230840252148396
+      "required_token_positions_per_second": 28617.659399074528,
+      "required_pages_per_second_upper_estimate": 0.002224897142497311
     },
     {
       "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": 28694.102433009662,
-      "required_pages_per_second_upper_estimate": 0.3680571366838504
+      "required_token_positions_per_second": 28617.659399074528,
+      "required_pages_per_second_upper_estimate": 0.3670766075226655
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12925941476567004,
-    "producer_p95_seconds": 0.19272138700034702,
-    "gpu_update_seconds": 0.963542946309196,
-    "producer_batches_per_second": 7.736380377497961,
-    "gpu_batches_per_second": 1.0378364595272591,
-    "producer_headroom": 7.454334742703036,
+    "producer_batch_seconds": 0.10201518258595854,
+    "producer_p95_seconds": 0.18670680400100537,
+    "gpu_update_seconds": 0.9661167468117297,
+    "producer_batches_per_second": 9.802462483046527,
+    "gpu_batches_per_second": 1.035071592848471,
+    "producer_headroom": 9.47032316486494,
     "sustainable": true,
-    "buffer_seconds": 60.70320561747935,
+    "buffer_seconds": 60.86535504913897,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327560.15754777315,
-    "upload_bytes_per_second": 4772732.680840122,
+    "generation_bytes_per_second": 342045.38893318584,
+    "upload_bytes_per_second": 4834421.459987242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.28678459954162,
+    "estimated_full_shards_per_hour": 73.39497805592234,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 992050,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.3297191130004649,
+      "bytes": 10315993,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -97,3 +97,4 @@
 {"step": 729700, "loss": 0.5729218125343323, "lr": 5e-05, "grad_norm": 0.22370433092843597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106807468726, "tokens_per_second": 152455.50771862615, "data_wait_s": 1.1952000932069495e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 729800, "loss": 0.5743268132209778, "lr": 5e-05, "grad_norm": 0.22477166533682702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106822037577, "tokens_per_second": 150901.01699446896, "data_wait_s": 1.8552998881204985e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 729900, "loss": 0.5633278489112854, "lr": 5e-05, "grad_norm": 0.2126331846159096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106836739730, "tokens_per_second": 152108.99362592484, "data_wait_s": 1.2638000043807551e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 730000, "loss": 0.5609837770462036, "lr": 5e-05, "grad_norm": 0.22295481463300923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106851433431, "tokens_per_second": 152340.38415329205, "data_wait_s": 8.98100006452296e-06, "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 sha256:634aa26e27b05a1731118deaf5764774817e5405e50e5553e81889074f48d5b3
-size 210630
+oid sha256:5c2957014f7c07c19c574f79f86dcd40b03f3e7e7370f3cff165a735a91f949d
+size 212748