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": 76810.37077788927,
-      "required_pages_per_second_upper_estimate": 0.005313919675029145
+      "required_token_positions_per_second": 77350.87323733325,
+      "required_pages_per_second_upper_estimate": 0.005351312889311971
     },
     {
       "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": 9601.296347236159,
-      "required_pages_per_second_upper_estimate": 0.004019862282720677
+      "required_token_positions_per_second": 9668.859154666656,
+      "required_pages_per_second_upper_estimate": 0.004048149419319971
     },
     {
       "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": 19202.592694472318,
-      "required_pages_per_second_upper_estimate": 0.003999393651834874
+      "required_token_positions_per_second": 19337.71830933331,
+      "required_pages_per_second_upper_estimate": 0.004027536753908319
     },
     {
       "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": 28803.889041708477,
-      "required_pages_per_second_upper_estimate": 0.0017267491670584386
+      "required_token_positions_per_second": 29006.577463999965,
+      "required_pages_per_second_upper_estimate": 0.0017389000284875123
     },
     {
       "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": 28803.889041708477,
-      "required_pages_per_second_upper_estimate": 0.002239375678074471
+      "required_token_positions_per_second": 29006.577463999965,
+      "required_pages_per_second_upper_estimate": 0.002255133811375485
     },
     {
       "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": 28803.889041708477,
-      "required_pages_per_second_upper_estimate": 0.3694653614205625
+      "required_token_positions_per_second": 29006.577463999965,
+      "required_pages_per_second_upper_estimate": 0.37206523087184573
     }
   ],
   "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": 0.11836015387507359,
-    "producer_p95_seconds": 0.33409919200494187,
-    "gpu_update_seconds": 0.9598703827793972,
-    "producer_batches_per_second": 8.448789286430609,
-    "gpu_batches_per_second": 1.0418073293442014,
-    "producer_headroom": 8.109742606388618,
+    "producer_batch_seconds": 0.11712624054706566,
+    "producer_p95_seconds": 0.3313373339988175,
+    "gpu_update_seconds": 0.9531631242711728,
+    "producer_batches_per_second": 8.5377964436429,
+    "gpu_batches_per_second": 1.049138363136573,
+    "producer_headroom": 8.137912732613975,
     "sustainable": true,
-    "buffer_seconds": 60.47183411510202,
+    "buffer_seconds": 60.04927682908389,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317069.09202244243,
-    "upload_bytes_per_second": 4980750.0573666515,
+    "generation_bytes_per_second": 303966.53793763486,
+    "upload_bytes_per_second": 4995073.260094661,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.03564615731196,
+    "estimated_full_shards_per_hour": 65.22414306256088,
     "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": 1.676982698001666,
-      "bytes": 6919911,
-      "pages": 22,
-      "shards": 3
-    },
     {
       "seconds": 1.0402027899981476,
       "bytes": 5156086,
@@ -562,6 +556,12 @@
       "bytes": 8697344,
       "pages": 49,
       "shards": 3
+    },
+    {
+      "seconds": 1.2496280530031072,
+      "bytes": 6028853,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -428,3 +428,4 @@
 {"step": 927800, "loss": 0.5746085047721863, "lr": 3e-05, "grad_norm": 0.3330007494133596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135803567729, "tokens_per_second": 127296.46218833102, "data_wait_s": 1.5518999134656042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927900, "loss": 0.5554463267326355, "lr": 3e-05, "grad_norm": 0.3334440388912399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135818387398, "tokens_per_second": 153709.56786431142, "data_wait_s": 1.2552998668979853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928000, "loss": 0.5243933200836182, "lr": 3e-05, "grad_norm": 0.33932875903643783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135833040179, "tokens_per_second": 153850.70217473313, "data_wait_s": 1.2272997992113233e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 928100, "loss": 0.55768883228302, "lr": 3e-05, "grad_norm": 0.35417443810619026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135847590762, "tokens_per_second": 151379.7995990841, "data_wait_s": 1.7454003682360053e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8de104b5213c3f23ba47a9b66e708178e560e15036a4925380a1f0c935dd6847
-size 917169
+oid sha256:960d0758766a447c79d4804d721c32f2c7c87dde559b2f67a4e2fa664f9ff8e1
+size 919287