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": 75035.11015588675,
-      "required_pages_per_second_upper_estimate": 0.0065038531735642354
+      "required_token_positions_per_second": 75343.61128962514,
+      "required_pages_per_second_upper_estimate": 0.006530593269947704
     },
     {
       "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": 9379.388769485844,
-      "required_pages_per_second_upper_estimate": 0.003741578414506879
+      "required_token_positions_per_second": 9417.951411203143,
+      "required_pages_per_second_upper_estimate": 0.003756961628850783
     },
     {
       "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": 18758.777538971688,
-      "required_pages_per_second_upper_estimate": 0.004339513040432205
+      "required_token_positions_per_second": 18835.902822406286,
+      "required_pages_per_second_upper_estimate": 0.00435735461739617
     },
     {
       "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": 28138.16630845753,
-      "required_pages_per_second_upper_estimate": 0.0017134666986358558
+      "required_token_positions_per_second": 28253.85423360943,
+      "required_pages_per_second_upper_estimate": 0.0017205114863099722
     },
     {
       "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": 28138.16630845753,
-      "required_pages_per_second_upper_estimate": 0.002166153933854676
+      "required_token_positions_per_second": 28253.85423360943,
+      "required_pages_per_second_upper_estimate": 0.002175059910577542
     },
     {
       "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": 28138.16630845753,
-      "required_pages_per_second_upper_estimate": 0.3493818531663401
+      "required_token_positions_per_second": 28253.85423360943,
+      "required_pages_per_second_upper_estimate": 0.35081831001414787
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04132590800794844,
-    "producer_p95_seconds": 0.1336728200003563,
-    "gpu_update_seconds": 0.982580019497923,
-    "producer_batches_per_second": 24.197895417268615,
-    "gpu_batches_per_second": 1.0177288161334466,
-    "producer_headroom": 23.776368550908497,
+    "producer_batch_seconds": 0.04043826431258424,
+    "producer_p95_seconds": 0.13931751500058454,
+    "gpu_update_seconds": 0.9785567580054713,
+    "producer_batches_per_second": 24.72905345961656,
+    "gpu_batches_per_second": 1.0219131305558966,
+    "producer_headroom": 24.198782381986362,
     "sustainable": true,
-    "buffer_seconds": 60.91996120887122,
+    "buffer_seconds": 60.67051899633922,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 63,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 85700, "loss": 0.6051968038082123, "lr": 0.0003, "grad_norm": 0.13198794077910492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12541770395, "tokens_per_second": 150425.11568733782, "data_wait_s": 1.1227000868530013e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 85800, "loss": 0.5887576583772898, "lr": 0.0003, "grad_norm": 0.16353512053089755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12556535738, "tokens_per_second": 151180.506023918, "data_wait_s": 8.628998330095783e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 85900, "loss": 0.6245568785816431, "lr": 0.0003, "grad_norm": 0.13522937998756454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12571121491, "tokens_per_second": 149098.87357553968, "data_wait_s": 9.809999028220773e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 86000, "loss": 0.5380822457373142, "lr": 0.0003, "grad_norm": 0.1317840326273743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12585673405, "tokens_per_second": 149026.47053418861, "data_wait_s": 7.379001544904895e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e1cd6a93159b42e39b107dbcd1f16611a1ea400a500c68e70007f772870437e
-size 19227
+oid sha256:4950490f945f162e3df9a507a1dbc155960feafa8feb70cf5fb01f33953dddce
+size 21345