Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75910.86578119127,
-      "required_pages_per_second_upper_estimate": 0.0065797614515832915
+      "required_token_positions_per_second": 76039.59732180713,
+      "required_pages_per_second_upper_estimate": 0.006590919575256767
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9488.858222648909,
-      "required_pages_per_second_upper_estimate": 0.0037852474160878047
+      "required_token_positions_per_second": 9504.949665225891,
+      "required_pages_per_second_upper_estimate": 0.0037916665331202692
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18977.716445297818,
-      "required_pages_per_second_upper_estimate": 0.004390160703217606
+      "required_token_positions_per_second": 19009.899330451783,
+      "required_pages_per_second_upper_estimate": 0.004397605647298543
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28466.574667946727,
-      "required_pages_per_second_upper_estimate": 0.001733465044703248
+      "required_token_positions_per_second": 28514.848995677672,
+      "required_pages_per_second_upper_estimate": 0.0017364046979862393
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28466.574667946727,
-      "required_pages_per_second_upper_estimate": 0.0021914357184606775
+      "required_token_positions_per_second": 28514.848995677672,
+      "required_pages_per_second_upper_estimate": 0.002195152009841298
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28466.574667946727,
-      "required_pages_per_second_upper_estimate": 0.35345958587912046
+      "required_token_positions_per_second": 28514.848995677672,
+      "required_pages_per_second_upper_estimate": 0.35405899146575703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05409720710139254,
-    "producer_p95_seconds": 0.19195511599900783,
-    "gpu_update_seconds": 0.9712443566711088,
-    "producer_batches_per_second": 18.48524265080329,
-    "gpu_batches_per_second": 1.029607012006175,
-    "producer_headroom": 17.95368760628878,
+    "producer_batch_seconds": 0.09894737310933976,
+    "producer_p95_seconds": 0.19850740400033828,
+    "gpu_update_seconds": 0.9696000846503142,
+    "producer_batches_per_second": 10.106382499866577,
+    "gpu_batches_per_second": 1.0313530452719066,
+    "producer_headroom": 9.799149327379086,
     "sustainable": true,
-    "buffer_seconds": 60.217150113608746,
+    "buffer_seconds": 60.115205248319484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291067.227243123,
-    "upload_bytes_per_second": 3750079.676134662,
+    "generation_bytes_per_second": 320547.4274874555,
+    "upload_bytes_per_second": 3734377.751770858,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.45625126810329,
+    "estimated_full_shards_per_hour": 68.78201597659825,
     "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": 1.4328443340000376,
-      "bytes": 11483713,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.909485917999973,
       "bytes": 1076059,
@@ -562,6 +556,12 @@
       "bytes": 6042311,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9439528090006206,
+      "bytes": 7900957,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -79,3 +79,4 @@
 {"step": 245000, "loss": 0.6086236331611872, "lr": 0.0003, "grad_norm": 0.15498698273521785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35860739606, "tokens_per_second": 148805.88176483728, "data_wait_s": 1.2215001333970577e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245100, "loss": 0.6002775095403194, "lr": 0.0003, "grad_norm": 0.16302515660379907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35875346044, "tokens_per_second": 137269.7995699142, "data_wait_s": 1.375700048811268e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 245200, "loss": 0.6231183186173439, "lr": 0.0003, "grad_norm": 0.1376267046021332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35889953945, "tokens_per_second": 149217.6129674794, "data_wait_s": 1.0816998837981373e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 245300, "loss": 0.6123337838798761, "lr": 0.0003, "grad_norm": 0.15098079689590116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35904726076, "tokens_per_second": 150987.21007849954, "data_wait_s": 1.6510999557795003e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:775a32c797f46863a87216d0172d818bc03aaaf1bec7e7a18b937db576b4aea5
-size 173289
+oid sha256:722ffd5838cfcaacba34110d25cc1c968761924dd61aa1c1c91f5dd3440b9b73
+size 175407