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": 76786.3947636734,
-      "required_pages_per_second_upper_estimate": 0.005312260958733672
+      "required_token_positions_per_second": 76080.28246898383,
+      "required_pages_per_second_upper_estimate": 0.005263410471780796
     },
     {
       "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": 9598.299345459174,
-      "required_pages_per_second_upper_estimate": 0.0040186075006611675
+      "required_token_positions_per_second": 9510.035308622979,
+      "required_pages_per_second_upper_estimate": 0.003981653191600535
     },
     {
       "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": 19196.59869091835,
-      "required_pages_per_second_upper_estimate": 0.0039981452589670854
+      "required_token_positions_per_second": 19020.070617245958,
+      "required_pages_per_second_upper_estimate": 0.003961379116579488
     },
     {
       "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": 28794.898036377523,
-      "required_pages_per_second_upper_estimate": 0.001726210170017318
+      "required_token_positions_per_second": 28530.105925868935,
+      "required_pages_per_second_upper_estimate": 0.0017103362873064705
     },
     {
       "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": 28794.898036377523,
-      "required_pages_per_second_upper_estimate": 0.0022386766669572427
+      "required_token_positions_per_second": 28530.105925868935,
+      "required_pages_per_second_upper_estimate": 0.002218090245062605
     },
     {
       "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": 28794.898036377523,
-      "required_pages_per_second_upper_estimate": 0.36935003445796644
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28530.105925868935,
+      "required_pages_per_second_upper_estimate": 0.36595356557597947
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10739503435956976,
-    "producer_p95_seconds": 0.3118640030006645,
-    "gpu_update_seconds": 0.9601700955867734,
-    "producer_batches_per_second": 9.31141747812935,
-    "gpu_batches_per_second": 1.0414821338388862,
-    "producer_headroom": 8.94054461002381,
+    "producer_batch_seconds": 0.13767621835916088,
+    "producer_p95_seconds": 0.3112715890019899,
+    "gpu_update_seconds": 0.9690815755062056,
+    "producer_batches_per_second": 7.263418562175089,
+    "gpu_batches_per_second": 1.0319048728974587,
+    "producer_headroom": 7.038845103793654,
     "sustainable": true,
-    "buffer_seconds": 60.49071602196672,
+    "buffer_seconds": 60.083057681384744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356530.4150557224,
-    "upload_bytes_per_second": 3781674.8128872192,
+    "generation_bytes_per_second": 341559.09278102586,
+    "upload_bytes_per_second": 3771164.695803873,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.5031274676681,
+    "estimated_full_shards_per_hour": 73.29063022206384,
     "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": 2.060914223002328,
-      "bytes": 10411136,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.9575544710023678,
       "bytes": 7322575,
@@ -562,6 +556,12 @@
       "bytes": 4595806,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.6076512380022905,
+      "bytes": 7510724,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 833500, "loss": 0.5352956056594849, "lr": 3e-05, "grad_norm": 0.2866640807895663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122001082959, "tokens_per_second": 152374.65815727136, "data_wait_s": 8.353999874088913e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 833600, "loss": 0.5327745676040649, "lr": 3e-05, "grad_norm": 0.29923131600335623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122015801724, "tokens_per_second": 152586.17669608677, "data_wait_s": 9.26199572859332e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 833700, "loss": 0.5276190638542175, "lr": 3e-05, "grad_norm": 0.30345125722342753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122030291745, "tokens_per_second": 149795.90730461536, "data_wait_s": 9.79100150289014e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 833800, "loss": 0.5325481295585632, "lr": 3e-05, "grad_norm": 0.3009460648840372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122044753261, "tokens_per_second": 149622.43569814428, "data_wait_s": 1.3773002137895674e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5df28b2e2c9de9e814fc31118b686db7f191f692f169190372196735c3eb2d71
-size 718629
+oid sha256:bbfe039fd839b146fc6f4a44f9e0c45ff23086ca5b66ea93be0d2af5e0a1ebf4
+size 720747