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": 77279.38132817368,
-      "required_pages_per_second_upper_estimate": 0.005346366913152229
+      "required_token_positions_per_second": 77632.6344389176,
+      "required_pages_per_second_upper_estimate": 0.005370805783013637
     },
     {
       "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": 9659.92266602171,
-      "required_pages_per_second_upper_estimate": 0.004044407898139436
+      "required_token_positions_per_second": 9704.0793048647,
+      "required_pages_per_second_upper_estimate": 0.004062895360727522
     },
     {
       "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": 19319.84533204342,
-      "required_pages_per_second_upper_estimate": 0.0040238142840809425
+      "required_token_positions_per_second": 19408.1586097294,
+      "required_pages_per_second_upper_estimate": 0.0040422076108451825
     },
     {
       "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": 28979.767998065134,
-      "required_pages_per_second_upper_estimate": 0.0017372928419404033
+      "required_token_positions_per_second": 29112.237914594094,
+      "required_pages_per_second_upper_estimate": 0.001745234211166464
     },
     {
       "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": 28979.767998065134,
-      "required_pages_per_second_upper_estimate": 0.00225304949332143
+      "required_token_positions_per_second": 29112.237914594094,
+      "required_pages_per_second_upper_estimate": 0.002263348446657973
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28979.767998065134,
-      "required_pages_per_second_upper_estimate": 0.3717213478285955
+      "required_token_positions_per_second": 29112.237914594094,
+      "required_pages_per_second_upper_estimate": 0.37342052968271433
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12178445642180691,
-    "producer_p95_seconds": 0.2773188860010123,
-    "gpu_update_seconds": 0.9540449047710096,
-    "producer_batches_per_second": 8.211228504698884,
-    "gpu_batches_per_second": 1.0481686920596474,
-    "producer_headroom": 7.833880716818446,
+    "producer_batch_seconds": 0.10995391027321944,
+    "producer_p95_seconds": 0.28422633899754146,
+    "gpu_update_seconds": 0.9497036978438519,
+    "producer_batches_per_second": 9.094719755897229,
+    "gpu_batches_per_second": 1.052959994017437,
+    "producer_headroom": 8.637288983029132,
     "sustainable": true,
-    "buffer_seconds": 60.1048290005736,
+    "buffer_seconds": 60.78103666200652,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480240.7900617476,
-    "upload_bytes_per_second": 4767325.286488481,
+    "generation_bytes_per_second": 613068.0702814669,
+    "upload_bytes_per_second": 4718575.914288704,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.04849411382027,
+    "estimated_full_shards_per_hour": 65.77506819406989,
     "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": 1.537398492997454,
-      "bytes": 7291338,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.5484525030005898,
       "bytes": 10816659,
@@ -562,6 +556,12 @@
       "bytes": 7495029,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.1257126100026653,
+      "bytes": 946737,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 917400, "loss": 0.540526807308197, "lr": 3e-05, "grad_norm": 0.3360390075632434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134280874026, "tokens_per_second": 154069.3524968311, "data_wait_s": 1.3517001207219437e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917500, "loss": 0.47982466220855713, "lr": 3e-05, "grad_norm": 0.3159958809459773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134295486274, "tokens_per_second": 152662.04811808284, "data_wait_s": 1.4806999388383701e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917600, "loss": 0.5721331834793091, "lr": 3e-05, "grad_norm": 0.3503418495413616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134310001711, "tokens_per_second": 151304.23800890677, "data_wait_s": 1.428399991709739e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 917700, "loss": 0.5132466554641724, "lr": 3e-05, "grad_norm": 0.32683448475066307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134324520310, "tokens_per_second": 151784.09833597168, "data_wait_s": 1.4648001524619758e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:052e7b051e2d0d26164d3f55ff0709eab68e026af7d05f3ab540832613f31fa9
-size 695331
+oid sha256:ee15477b2dcf68b24bece49c769ee33f0f33797785ec39f8d8cb22dc2df0a115
+size 697449