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": 77085.93677079078,
-      "required_pages_per_second_upper_estimate": 0.005332983969819267
+      "required_token_positions_per_second": 76064.73353233763,
+      "required_pages_per_second_upper_estimate": 0.005262334760265109
     },
     {
       "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": 9635.742096348848,
-      "required_pages_per_second_upper_estimate": 0.004034283998565123
+      "required_token_positions_per_second": 9508.091691542204,
+      "required_pages_per_second_upper_estimate": 0.003980839439716154
     },
     {
       "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": 19271.484192697695,
-      "required_pages_per_second_upper_estimate": 0.0040137419341242375
+      "required_token_positions_per_second": 19016.183383084408,
+      "required_pages_per_second_upper_estimate": 0.003960569508216896
     },
     {
       "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": 28907.226289046543,
-      "required_pages_per_second_upper_estimate": 0.001732944077249515
+      "required_token_positions_per_second": 28524.275074626614,
+      "required_pages_per_second_upper_estimate": 0.0017099867366776875
     },
     {
       "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": 28907.226289046543,
-      "required_pages_per_second_upper_estimate": 0.0022474096945224913
+      "required_token_positions_per_second": 28524.275074626614,
+      "required_pages_per_second_upper_estimate": 0.0022176369220257186
     },
     {
       "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": 28907.226289046543,
-      "required_pages_per_second_upper_estimate": 0.3707908606745237
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28524.275074626614,
+      "required_pages_per_second_upper_estimate": 0.36587877367692323
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05990864222650316,
-    "producer_p95_seconds": 0.215352298999278,
-    "gpu_update_seconds": 0.9564390482692666,
-    "producer_batches_per_second": 16.69208252490835,
-    "gpu_batches_per_second": 1.0455449323295192,
-    "producer_headroom": 15.9649595237554,
+    "producer_batch_seconds": 0.08341054327343045,
+    "producer_p95_seconds": 0.21766731900061131,
+    "gpu_update_seconds": 0.9692796724076577,
+    "producer_batches_per_second": 11.988892060346279,
+    "gpu_batches_per_second": 1.0316939769468538,
+    "producer_headroom": 11.62058936878321,
     "sustainable": true,
-    "buffer_seconds": 60.2556600409638,
+    "buffer_seconds": 60.09533968927477,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452548.80088120635,
-    "upload_bytes_per_second": 5065018.889223543,
+    "generation_bytes_per_second": 460472.4781085898,
+    "upload_bytes_per_second": 5016190.616838258,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.10643787219183,
+    "estimated_full_shards_per_hour": 98.8066745514228,
     "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": 1.150645027999417,
-      "bytes": 6104360,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1603049650002504,
       "bytes": 1160533,
@@ -562,6 +556,12 @@
       "bytes": 12329538,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 0.9383213960009016,
+      "bytes": 933737,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -157,3 +157,4 @@
 {"step": 455900, "loss": 0.5676047801971436, "lr": 0.0003, "grad_norm": 0.16011803222362866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66729768871, "tokens_per_second": 151271.31900934165, "data_wait_s": 7.996000931598246e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 456000, "loss": 0.5559360384941101, "lr": 0.0003, "grad_norm": 0.19572494407885363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66744473858, "tokens_per_second": 153701.48077976104, "data_wait_s": 1.487400004407391e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456100, "loss": 0.5706086754798889, "lr": 0.0003, "grad_norm": 0.14132408978913322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66759121405, "tokens_per_second": 152543.7970543519, "data_wait_s": 1.6601999959675595e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 456200, "loss": 0.5912990570068359, "lr": 0.0003, "grad_norm": 0.1895758592389742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66773778104, "tokens_per_second": 152334.10924017965, "data_wait_s": 1.1068001185776666e-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 sha256:20eb6b462ea1e7fd9d65b5ed2e037075e7aaba3036dcad19e06f6cf7853e5185
-size 339276
+oid sha256:7ebd2003eb64d8f66886047d041a5fb71031c392ddd0b7af5b57d72e85fbd82c
+size 341394