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": 76675.74287650445,
-      "required_pages_per_second_upper_estimate": 0.005304605804431574
+      "required_token_positions_per_second": 77223.5356475711,
+      "required_pages_per_second_upper_estimate": 0.005342503379388311
     },
     {
       "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": 9584.467859563056,
-      "required_pages_per_second_upper_estimate": 0.004012816546350733
+      "required_token_positions_per_second": 9652.941955946388,
+      "required_pages_per_second_upper_estimate": 0.004041485220604701
     },
     {
       "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": 19168.93571912611,
-      "required_pages_per_second_upper_estimate": 0.003992383791464387
+      "required_token_positions_per_second": 19305.883911892775,
+      "required_pages_per_second_upper_estimate": 0.004020906488450973
     },
     {
       "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": 28753.403578689165,
-      "required_pages_per_second_upper_estimate": 0.0017237226406372694
+      "required_token_positions_per_second": 28958.82586783916,
+      "required_pages_per_second_upper_estimate": 0.0017360373932101378
     },
     {
       "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": 28753.403578689165,
-      "required_pages_per_second_upper_estimate": 0.0022354506553868034
+      "required_token_positions_per_second": 28958.82586783916,
+      "required_pages_per_second_upper_estimate": 0.0022514213348110565
     },
     {
       "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": 28753.403578689165,
-      "required_pages_per_second_upper_estimate": 0.3688177881080177
+      "required_token_positions_per_second": 28958.82586783916,
+      "required_pages_per_second_upper_estimate": 0.37145272466796425
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3397120472498898,
-    "producer_p95_seconds": 0.2708103039985872,
-    "gpu_update_seconds": 0.9615557311097443,
-    "producer_batches_per_second": 2.943669522748503,
-    "gpu_batches_per_second": 1.0399813215671718,
-    "producer_headroom": 2.830502300091909,
+    "producer_batch_seconds": 0.0995859719061798,
+    "producer_p95_seconds": 0.280140177001158,
+    "gpu_update_seconds": 0.9547348406381722,
+    "producer_batches_per_second": 10.041574941319071,
+    "gpu_batches_per_second": 1.0474112365393216,
+    "producer_headroom": 9.587041451356527,
     "sustainable": true,
-    "buffer_seconds": 60.57801105991389,
+    "buffer_seconds": 60.148294960204844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432816.2816446204,
-    "upload_bytes_per_second": 4850650.674575459,
+    "generation_bytes_per_second": 417538.79051506787,
+    "upload_bytes_per_second": 4896569.697884343,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.87229859355888,
+    "estimated_full_shards_per_hour": 89.59410463894869,
     "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": 1.6746519440002885,
-      "bytes": 10811116,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 1.732792734999748,
       "bytes": 6311900,
@@ -562,6 +556,12 @@
       "bytes": 6019433,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.3825668060017051,
+      "bytes": 13491032,
+      "pages": 40,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -239,3 +239,4 @@
 {"step": 908900, "loss": 0.5288899540901184, "lr": 3e-05, "grad_norm": 0.32810802951407375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133036892421, "tokens_per_second": 152041.56680291082, "data_wait_s": 1.4968001778470352e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909000, "loss": 0.5073631405830383, "lr": 3e-05, "grad_norm": 0.3221906008646696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133051544757, "tokens_per_second": 153064.6997042795, "data_wait_s": 1.0700001439545304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909100, "loss": 0.5431069135665894, "lr": 3e-05, "grad_norm": 0.3196008440782156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133066160939, "tokens_per_second": 151945.5474734111, "data_wait_s": 1.3926000974606723e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 909200, "loss": 0.6002414226531982, "lr": 3e-05, "grad_norm": 0.3585365651129577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133080864528, "tokens_per_second": 153427.0324208804, "data_wait_s": 1.3517001207219437e-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 sha256:93ae098602365d5fbe06353b24d1e0e72ac49eb7fd55e38ade2d314ce82e4f4a
-size 513735
+oid sha256:9a743b136ea6f226144ae6fe94f632c0cef8780b801d6c6e1e40c12c4ba88fcd
+size 515853