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": 76664.37535911164,
-      "required_pages_per_second_upper_estimate": 0.005303819373202065
+      "required_token_positions_per_second": 74758.47360337824,
+      "required_pages_per_second_upper_estimate": 0.005171964667439081
     },
     {
       "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": 9583.046919888955,
-      "required_pages_per_second_upper_estimate": 0.004012221628581781
+      "required_token_positions_per_second": 9344.80920042228,
+      "required_pages_per_second_upper_estimate": 0.003912476470410389
     },
     {
       "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": 19166.09383977791,
-      "required_pages_per_second_upper_estimate": 0.003991791902941555
+      "required_token_positions_per_second": 18689.61840084456,
+      "required_pages_per_second_upper_estimate": 0.0038925546345140558
     },
     {
       "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": 28749.140759666865,
-      "required_pages_per_second_upper_estimate": 0.0017234670911458303
+      "required_token_positions_per_second": 28034.42760126684,
+      "required_pages_per_second_upper_estimate": 0.0016806211286035531
     },
     {
       "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": 28749.140759666865,
-      "required_pages_per_second_upper_estimate": 0.002235119239957979
+      "required_token_positions_per_second": 28034.42760126684,
+      "required_pages_per_second_upper_estimate": 0.0021795534355833254
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28749.140759666865,
-      "required_pages_per_second_upper_estimate": 0.36876310924265804
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28034.42760126684,
+      "required_pages_per_second_upper_estimate": 0.35959553624590296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05364367957032101,
-    "producer_p95_seconds": 0.1745841570000266,
-    "gpu_update_seconds": 0.9616983071295232,
-    "producer_batches_per_second": 18.641525115537778,
-    "gpu_batches_per_second": 1.0398271397448953,
-    "producer_headroom": 17.92752314592517,
+    "producer_batch_seconds": 0.4241352658438089,
+    "producer_p95_seconds": 0.17352022499926534,
+    "gpu_update_seconds": 0.9862159625028556,
+    "producer_batches_per_second": 2.357738392751943,
+    "gpu_batches_per_second": 1.0139766927541538,
+    "producer_headroom": 2.325239238337793,
     "sustainable": true,
-    "buffer_seconds": 60.58699334915996,
+    "buffer_seconds": 60.15917371267419,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333379.6292513174,
-    "upload_bytes_per_second": 4722995.010686704,
+    "generation_bytes_per_second": 302739.74470146885,
+    "upload_bytes_per_second": 4562079.956788493,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.53550775675431,
+    "estimated_full_shards_per_hour": 64.9609017923646,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 6054602,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0668450510002003,
+      "bytes": 1106491,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -61,3 +61,4 @@
 {"step": 726100, "loss": 0.5922380089759827, "lr": 5e-05, "grad_norm": 0.24529023929439142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106280761074, "tokens_per_second": 150243.22403892863, "data_wait_s": 1.6334000065398868e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726200, "loss": 0.5506969094276428, "lr": 5e-05, "grad_norm": 0.2236093842998674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106295367759, "tokens_per_second": 151224.27904470032, "data_wait_s": 1.4868000107526314e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726300, "loss": 0.5189372301101685, "lr": 5e-05, "grad_norm": 0.2189112038765758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106310085132, "tokens_per_second": 151529.8049592878, "data_wait_s": 8.615000297140796e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 726400, "loss": 0.5483208298683167, "lr": 5e-05, "grad_norm": 0.2387842533931959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106324813331, "tokens_per_second": 151435.70127050293, "data_wait_s": 1.1844000255223364e-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 sha256:398d9b14598f116b32ca14c27a594dd5da92d0237dc5928ca6ef5ffbf3aa4b9a
-size 134382
+oid sha256:b291e6dd19c6b5f01a4e2e7b53871f3a1ac3e2910bedc2445ae4b5a91d2f4de4
+size 136500