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": 76155.301767764,
-      "required_pages_per_second_upper_estimate": 0.005268600480939164
+      "required_token_positions_per_second": 76109.47044479584,
+      "required_pages_per_second_upper_estimate": 0.005265429763672906
     },
     {
       "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": 9519.4127209705,
-      "required_pages_per_second_upper_estimate": 0.003985579318327804
+      "required_token_positions_per_second": 9513.68380559948,
+      "required_pages_per_second_upper_estimate": 0.003983180741095315
     },
     {
       "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": 19038.825441941,
-      "required_pages_per_second_upper_estimate": 0.003965285251965479
+      "required_token_positions_per_second": 19027.36761119896,
+      "required_pages_per_second_upper_estimate": 0.003962898887985227
     },
     {
       "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": 28558.238162911504,
-      "required_pages_per_second_upper_estimate": 0.0017120227719617338
+      "required_token_positions_per_second": 28541.051416798437,
+      "required_pages_per_second_upper_estimate": 0.0017109924527748972
     },
     {
       "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": 28558.238162911504,
-      "required_pages_per_second_upper_estimate": 0.0022202774027520315
+      "required_token_positions_per_second": 28541.051416798437,
+      "required_pages_per_second_upper_estimate": 0.0022189412088382474
     },
     {
       "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": 28558.238162911504,
-      "required_pages_per_second_upper_estimate": 0.3663144157067188
+      "required_token_positions_per_second": 28541.051416798437,
+      "required_pages_per_second_upper_estimate": 0.3660939625812706
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0972743232500477,
-    "producer_p95_seconds": 0.28578925000329036,
-    "gpu_update_seconds": 0.9681269496486787,
-    "producer_batches_per_second": 10.280205161946574,
-    "gpu_batches_per_second": 1.032922387258084,
-    "producer_headroom": 9.952543665197938,
+    "producer_batch_seconds": 0.07955510920314168,
+    "producer_p95_seconds": 0.2891347170007066,
+    "gpu_update_seconds": 0.9687099327997142,
+    "producer_batches_per_second": 12.569902926618186,
+    "gpu_batches_per_second": 1.0323007601561933,
+    "producer_headroom": 12.176589819343233,
     "sustainable": true,
-    "buffer_seconds": 60.023870878218084,
+    "buffer_seconds": 60.06001583358228,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 562588.3340588472,
-    "upload_bytes_per_second": 5012517.76447907,
+    "generation_bytes_per_second": 566351.1648585552,
+    "upload_bytes_per_second": 4989871.069013539,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.359180051441484,
+    "estimated_full_shards_per_hour": 60.76288799914118,
     "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": 1.0521548180004174,
-      "bytes": 6267595,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9623302439977124,
       "bytes": 6111937,
@@ -562,6 +556,12 @@
       "bytes": 13269645,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.3757806130015524,
+      "bytes": 5911161,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -251,3 +251,4 @@
 {"step": 465300, "loss": 0.570895791053772, "lr": 0.0003, "grad_norm": 0.16260643659947774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68105681035, "tokens_per_second": 150759.28425460175, "data_wait_s": 8.406997949350625e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 465400, "loss": 0.5856493711471558, "lr": 0.0003, "grad_norm": 0.17388075203956013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68120467123, "tokens_per_second": 153737.55892069484, "data_wait_s": 1.8632999854162335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465500, "loss": 0.6045053601264954, "lr": 0.0003, "grad_norm": 0.17046018346913602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68135156367, "tokens_per_second": 152315.42781032363, "data_wait_s": 1.2090997188352048e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 465600, "loss": 0.6022682785987854, "lr": 0.0003, "grad_norm": 0.1557973367140862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68149959809, "tokens_per_second": 153728.81772058288, "data_wait_s": 1.483599771745503e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b6adbc11b530313e39d2fb2ab4dcb1ce7ad7fed041d94280ccf132ff220befd
-size 539934
+oid sha256:4ff7f4ddfbc97c744ff93ccd374e54fcb7a212af81da240f16f912d85c37cb67
+size 542052