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": 77586.51064115175,
-      "required_pages_per_second_upper_estimate": 0.00536761483166224
+      "required_token_positions_per_second": 76860.04282470059,
+      "required_pages_per_second_upper_estimate": 0.005317356102482579
     },
     {
       "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": 9698.313830143969,
-      "required_pages_per_second_upper_estimate": 0.004060481476858754
+      "required_token_positions_per_second": 9607.505353087574,
+      "required_pages_per_second_upper_estimate": 0.0040224618638118785
     },
     {
       "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": 19396.627660287937,
-      "required_pages_per_second_upper_estimate": 0.004039806018168112
+      "required_token_positions_per_second": 19215.010706175148,
+      "required_pages_per_second_upper_estimate": 0.004001979996187582
     },
     {
       "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": 29094.941490431906,
-      "required_pages_per_second_upper_estimate": 0.0017441973169480512
+      "required_token_positions_per_second": 28822.51605926272,
+      "required_pages_per_second_upper_estimate": 0.0017278658283190066
     },
     {
       "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": 29094.941490431906,
-      "required_pages_per_second_upper_estimate": 0.0022620037257582923
+      "required_token_positions_per_second": 28822.51605926272,
+      "required_pages_per_second_upper_estimate": 0.0022408238467577224
     },
     {
       "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": 29094.941490431906,
-      "required_pages_per_second_upper_estimate": 0.37319866972501514
+      "required_token_positions_per_second": 28822.51605926272,
+      "required_pages_per_second_upper_estimate": 0.3697042888016152
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13600141748452188,
+    "producer_batch_seconds": 0.10882085524173135,
     "producer_p95_seconds": 0.3024372569998377,
-    "gpu_update_seconds": 0.9502682797658231,
-    "producer_batches_per_second": 7.352864539914141,
-    "gpu_batches_per_second": 1.0523343999722188,
-    "producer_headroom": 6.987193937695331,
+    "gpu_update_seconds": 0.9592500510070748,
+    "producer_batches_per_second": 9.189415004859411,
+    "gpu_batches_per_second": 1.042481049597176,
+    "producer_headroom": 8.814946812136569,
     "sustainable": true,
-    "buffer_seconds": 60.817169905012676,
+    "buffer_seconds": 60.432753213445714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360227.78244158806,
-    "upload_bytes_per_second": 5111304.922121814,
+    "generation_bytes_per_second": 359308.43726154964,
+    "upload_bytes_per_second": 4889450.748332223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.29649643836719,
+    "estimated_full_shards_per_hour": 77.09922636399142,
     "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": 1.4434581130008155,
-      "bytes": 11762868,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.317350670000451,
       "bytes": 2491706,
@@ -562,6 +556,12 @@
       "bytes": 6248444,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 3.389049104996957,
+      "bytes": 2610225,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -362,3 +362,4 @@
 {"step": 921200, "loss": 0.5252535939216614, "lr": 3e-05, "grad_norm": 0.3546703675861652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134837222360, "tokens_per_second": 153886.84473517977, "data_wait_s": 1.3698998373001814e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921300, "loss": 0.5563081502914429, "lr": 3e-05, "grad_norm": 0.3281522780104051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134852016869, "tokens_per_second": 154609.68895645038, "data_wait_s": 1.8147999071516097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921400, "loss": 0.5326442718505859, "lr": 3e-05, "grad_norm": 0.33162075839769756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134866552710, "tokens_per_second": 151936.60879466034, "data_wait_s": 1.3748998753726482e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 921500, "loss": 0.5057809948921204, "lr": 3e-05, "grad_norm": 0.3358195532381215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134881069707, "tokens_per_second": 152151.62702244002, "data_wait_s": 1.3453995052259415e-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 sha256:88fed15478dd5899f728f1bf4ffd268469c5912a24ddda1f146d2f74c01b7a2c
-size 776598
+oid sha256:cb2c6e9fb4eace2553d813b7392958d963f59a0ceae1ccd5aade0f6e940ddf6f
+size 778716