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": 77151.40965522398,
-      "required_pages_per_second_upper_estimate": 0.005337513535882378
+      "required_token_positions_per_second": 74929.07690904371,
+      "required_pages_per_second_upper_estimate": 0.005183767399979225
     },
     {
       "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": 9643.926206902997,
-      "required_pages_per_second_upper_estimate": 0.004037710514750483
+      "required_token_positions_per_second": 9366.134613630464,
+      "required_pages_per_second_upper_estimate": 0.003921404975595388
     },
     {
       "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": 19287.852413805995,
-      "required_pages_per_second_upper_estimate": 0.0040171510029220776
+      "required_token_positions_per_second": 18732.269227260927,
+      "required_pages_per_second_upper_estimate": 0.0039014376768786545
     },
     {
       "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": 28931.778620708992,
-      "required_pages_per_second_upper_estimate": 0.001734415952043439
+      "required_token_positions_per_second": 28098.40384089139,
+      "required_pages_per_second_upper_estimate": 0.0016844564064829833
     },
     {
       "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": 28931.778620708992,
-      "required_pages_per_second_upper_estimate": 0.002249318530314953
+      "required_token_positions_per_second": 28098.40384089139,
+      "required_pages_per_second_upper_estimate": 0.0021845273068123953
     },
     {
       "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": 28931.778620708992,
-      "required_pages_per_second_upper_estimate": 0.3711057916228498
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28098.40384089139,
+      "required_pages_per_second_upper_estimate": 0.3604161547554725
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0800984749294571,
-    "producer_p95_seconds": 0.28514181100035785,
-    "gpu_update_seconds": 0.9556273868420215,
-    "producer_batches_per_second": 12.484632209048952,
-    "gpu_batches_per_second": 1.0464329651587454,
-    "producer_headroom": 11.930656453617184,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.20452537104735,
+    "buffer_seconds": 60.0221994654945,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 500311.0840473855,
+    "generation_bytes_per_second": 450691.244022836,
     "upload_bytes_per_second": 5111904.656332825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.35511199060606,
+    "estimated_full_shards_per_hour": 96.70784941209611,
     "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": [

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

@@ -264,3 +264,4 @@
 {"step": 466600, "loss": 0.5827690958976746, "lr": 0.0003, "grad_norm": 0.17281665825381523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68296585906, "tokens_per_second": 153156.93834537195, "data_wait_s": 1.5769001038279384e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:23b5eb2f408015093c57d8952168cf90df666e4cc0a3a3d7fbb1f206fbac2ab5
-size 567468
+oid sha256:66a5772099c619b75ddb9c7ae15253045793c505e7cdb4fc96aa9a5ce0617740
+size 569586