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": 75202.94144948688,
-      "required_pages_per_second_upper_estimate": 0.0052027139843403665
+      "required_token_positions_per_second": 77175.47242258527,
+      "required_pages_per_second_upper_estimate": 0.005339178253961744
     },
     {
       "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": 9400.36768118586,
-      "required_pages_per_second_upper_estimate": 0.003935737646113092
+      "required_token_positions_per_second": 9646.934052823159,
+      "required_pages_per_second_upper_estimate": 0.004038969837026289
     },
     {
       "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": 18800.73536237172,
-      "required_pages_per_second_upper_estimate": 0.00391569736724883
+      "required_token_positions_per_second": 19293.868105646317,
+      "required_pages_per_second_upper_estimate": 0.004018403912887955
     },
     {
       "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": 28201.10304355758,
-      "required_pages_per_second_upper_estimate": 0.0016906130668702187
+      "required_token_positions_per_second": 28940.802158469476,
+      "required_pages_per_second_upper_estimate": 0.0017349568993540359
     },
     {
       "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": 28201.10304355758,
-      "required_pages_per_second_upper_estimate": 0.002192511718093639
+      "required_token_positions_per_second": 28940.802158469476,
+      "required_pages_per_second_upper_estimate": 0.002250020070685483
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28201.10304355758,
-      "required_pages_per_second_upper_estimate": 0.3617334698574618
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28940.802158469476,
+      "required_pages_per_second_upper_estimate": 0.3712215358765213
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4949295265859064,
-    "producer_p95_seconds": 0.36481483899842715,
-    "gpu_update_seconds": 0.9803871840508048,
-    "producer_batches_per_second": 2.02048967839551,
-    "gpu_batches_per_second": 1.0200051737397853,
-    "producer_headroom": 1.9808621862058904,
+    "producer_batch_seconds": 0.14964189717233012,
+    "producer_p95_seconds": 0.37099440700694686,
+    "gpu_update_seconds": 0.9553294289705394,
+    "producer_batches_per_second": 6.682620435160503,
+    "gpu_batches_per_second": 1.046759337328902,
+    "producer_headroom": 6.384103964348741,
     "sustainable": true,
-    "buffer_seconds": 60.7840054111499,
+    "buffer_seconds": 60.185754025143986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296743.85111903306,
-    "upload_bytes_per_second": 4521540.499346272,
+    "generation_bytes_per_second": 298859.1830600483,
+    "upload_bytes_per_second": 4467479.035478239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.674322606832924,
+    "estimated_full_shards_per_hour": 64.12822359896744,
     "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": 1.5149503700013156,
-      "bytes": 16156661,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 1.2590785899956245,
       "bytes": 6075199,
@@ -562,6 +556,12 @@
       "bytes": 6156537,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6308916679990944,
+      "bytes": 11465498,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -422,3 +422,4 @@
 {"step": 646600, "loss": 0.5488482117652893, "lr": 0.0001, "grad_norm": 0.1857746478974556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94645761373, "tokens_per_second": 153547.3558024513, "data_wait_s": 1.0648996976669878e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646700, "loss": 0.5114997029304504, "lr": 0.0001, "grad_norm": 0.17141576251459942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94660486978, "tokens_per_second": 152009.3076243558, "data_wait_s": 1.0433002898935229e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646800, "loss": 0.5721449255943298, "lr": 0.0001, "grad_norm": 0.17860234885716397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94675145228, "tokens_per_second": 151517.47726566327, "data_wait_s": 1.3228003808762878e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 646900, "loss": 0.557778000831604, "lr": 0.0001, "grad_norm": 0.1944920305595768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94689630493, "tokens_per_second": 149477.76799090422, "data_wait_s": 1.7053003830369562e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c067602fda7c6a89895758cb447373776a1f1cb7c69a355db6c86550f202c250
-size 905244
+oid sha256:8b8c850ebb3ec20541b863ace42502ba49c8f89ba56438edd2be7f6631c766ad
+size 907362