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": 74929.07690904371,
-      "required_pages_per_second_upper_estimate": 0.005183767399979225
+      "required_token_positions_per_second": 76166.63031477894,
+      "required_pages_per_second_upper_estimate": 0.005269384216107508
     },
     {
       "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": 9366.134613630464,
-      "required_pages_per_second_upper_estimate": 0.003921404975595388
+      "required_token_positions_per_second": 9520.828789347368,
+      "required_pages_per_second_upper_estimate": 0.0039861721965863284
     },
     {
       "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": 18732.269227260927,
-      "required_pages_per_second_upper_estimate": 0.0039014376768786545
+      "required_token_positions_per_second": 19041.657578694736,
+      "required_pages_per_second_upper_estimate": 0.003965875111362813
     },
     {
       "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": 28098.40384089139,
-      "required_pages_per_second_upper_estimate": 0.0016844564064829833
+      "required_token_positions_per_second": 28562.486368042104,
+      "required_pages_per_second_upper_estimate": 0.0017122774453726796
     },
     {
       "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": 28098.40384089139,
-      "required_pages_per_second_upper_estimate": 0.0021845273068123953
+      "required_token_positions_per_second": 28562.486368042104,
+      "required_pages_per_second_upper_estimate": 0.002220607682015056
     },
     {
       "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": 28098.40384089139,
-      "required_pages_per_second_upper_estimate": 0.3604161547554725
+      "required_token_positions_per_second": 28562.486368042104,
+      "required_pages_per_second_upper_estimate": 0.36636890712076686
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4413017245782953,
-    "producer_p95_seconds": 0.3069951540019247,
-    "gpu_update_seconds": 0.9839704830408935,
-    "producer_batches_per_second": 2.266023322151285,
-    "gpu_batches_per_second": 1.0162906481803888,
-    "producer_headroom": 2.2297000628791297,
+    "producer_batch_seconds": 0.1038662158826753,
+    "producer_p95_seconds": 0.2904386770023848,
+    "gpu_update_seconds": 0.9679829565165132,
+    "producer_batches_per_second": 9.627769640992554,
+    "gpu_batches_per_second": 1.033076040510782,
+    "producer_headroom": 9.319516921747901,
     "sustainable": true,
-    "buffer_seconds": 60.0221994654945,
+    "buffer_seconds": 60.01494330402382,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450691.244022836,
-    "upload_bytes_per_second": 5111904.656332825,
+    "generation_bytes_per_second": 450558.9683641795,
+    "upload_bytes_per_second": 5255600.4674839545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.70784941209611,
+    "estimated_full_shards_per_hour": 96.67946613496817,
     "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": 0.9373651969981438,
-      "bytes": 2246840,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 2.0274646729994856,
       "bytes": 10144494,
@@ -562,6 +556,12 @@
       "bytes": 6207129,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1704257509991294,
+      "bytes": 15475495,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -265,3 +265,4 @@
 {"step": 466700, "loss": 0.5531118512153625, "lr": 0.0003, "grad_norm": 0.14978028778130684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68311272192, "tokens_per_second": 152311.2333625329, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466800, "loss": 0.5992187261581421, "lr": 0.0003, "grad_norm": 0.16698154940396948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68325949648, "tokens_per_second": 152670.00531683842, "data_wait_s": 1.3724999007536098e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466900, "loss": 0.6337512731552124, "lr": 0.0003, "grad_norm": 0.1660150600172574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68340525764, "tokens_per_second": 150606.48838295694, "data_wait_s": 1.109000004362315e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 467000, "loss": 0.6555135846138, "lr": 0.0003, "grad_norm": 0.23605228265184078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68355048751, "tokens_per_second": 150920.1984837939, "data_wait_s": 1.2935000995639712e-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 sha256:66a5772099c619b75ddb9c7ae15253045793c505e7cdb4fc96aa9a5ce0617740
-size 569586
+oid sha256:5300c7b9b355feeffa00e521f1c50865ec8833d7b878a7e9b9526a3fb9dbde9c
+size 571704