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": 75102.22448560117,
-      "required_pages_per_second_upper_estimate": 0.0065096704735641674
+      "required_token_positions_per_second": 75495.7471401701,
+      "required_pages_per_second_upper_estimate": 0.0065437800198881535
     },
     {
       "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": 9387.778060700146,
-      "required_pages_per_second_upper_estimate": 0.003744925028203345
+      "required_token_positions_per_second": 9436.968392521263,
+      "required_pages_per_second_upper_estimate": 0.0037645477870277898
     },
     {
       "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": 18775.556121400292,
-      "required_pages_per_second_upper_estimate": 0.004343394470184097
+      "required_token_positions_per_second": 18873.936785042526,
+      "required_pages_per_second_upper_estimate": 0.004366153105277184
     },
     {
       "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": 28163.334182100436,
-      "required_pages_per_second_upper_estimate": 0.0017149992900950808
+      "required_token_positions_per_second": 28310.905177563785,
+      "required_pages_per_second_upper_estimate": 0.0017239855894736224
     },
     {
       "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": 28163.334182100436,
-      "required_pages_per_second_upper_estimate": 0.00216809142643684
+      "required_token_positions_per_second": 28310.905177563785,
+      "required_pages_per_second_upper_estimate": 0.0021794518501702695
     },
     {
       "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": 28163.334182100436,
-      "required_pages_per_second_upper_estimate": 0.34969435392553033
+      "required_token_positions_per_second": 28310.905177563785,
+      "required_pages_per_second_upper_estimate": 0.35152669180083423
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.14321208446861533,
-    "producer_p95_seconds": 0.37185477900129627,
-    "gpu_update_seconds": 0.9817019469794183,
-    "producer_batches_per_second": 6.982650966295712,
-    "gpu_batches_per_second": 1.0186391124891652,
-    "producer_headroom": 6.854882048690217,
+    "producer_batch_seconds": 0.1081305857735515,
+    "producer_p95_seconds": 0.3642636240001593,
+    "gpu_update_seconds": 0.976584811633323,
+    "producer_batches_per_second": 9.24807715454546,
+    "gpu_batches_per_second": 1.0239766050912829,
+    "producer_headroom": 9.031531685942216,
     "sustainable": true,
-    "buffer_seconds": 60.865520712723935,
+    "buffer_seconds": 60.548258321266026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269297.7603940833,
-    "upload_bytes_per_second": 5295981.193420068,
+    "generation_bytes_per_second": 268596.76830957545,
+    "upload_bytes_per_second": 5353718.983507786,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.78503044955133,
+    "estimated_full_shards_per_hour": 57.63461386647651,
     "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": 1.516586848996667,
-      "bytes": 2356968,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1747013890017115,
       "bytes": 10632168,
@@ -562,6 +556,12 @@
       "bytes": 7350483,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1773676480006543,
+      "bytes": 5012862,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 226200, "loss": 0.5835002064704895, "lr": 0.0003, "grad_norm": 0.13729538664677776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33109675151, "tokens_per_second": 150500.3309403648, "data_wait_s": 1.211600101669319e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226300, "loss": 0.5875837076455355, "lr": 0.0003, "grad_norm": 0.1494478455169779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33124224171, "tokens_per_second": 147774.17393377807, "data_wait_s": 1.5002999134594575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226400, "loss": 0.5998847484588623, "lr": 0.0003, "grad_norm": 0.14292261352133273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33138742060, "tokens_per_second": 147449.55271050605, "data_wait_s": 2.0502997358562425e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 226500, "loss": 0.6257619615644217, "lr": 0.0003, "grad_norm": 0.15677036568711947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33153327328, "tokens_per_second": 148919.01146690917, "data_wait_s": 9.450999641558155e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a724808b4e2bbfb38588e89d545962c2029fa4295a7b861f4d12824123a19bd
-size 669915
+oid sha256:f1b3e8657e96d5d23f39b1cd56be6bfefe8e9fc05ed601988fd27c124a6c555e
+size 672033