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": 75475.41993424475,
-      "required_pages_per_second_upper_estimate": 0.006542018109197376
+      "required_token_positions_per_second": 75877.688418319,
+      "required_pages_per_second_upper_estimate": 0.00657688572185678
     },
     {
       "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": 9434.427491780594,
-      "required_pages_per_second_upper_estimate": 0.003763534183732485
+      "required_token_positions_per_second": 9484.711052289875,
+      "required_pages_per_second_upper_estimate": 0.0037835930478258635
     },
     {
       "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": 18868.85498356119,
-      "required_pages_per_second_upper_estimate": 0.004364977519941128
+      "required_token_positions_per_second": 18969.42210457975,
+      "required_pages_per_second_upper_estimate": 0.0043882419534678925
     },
     {
       "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": 28303.282475341784,
-      "required_pages_per_second_upper_estimate": 0.0017235214069015303
+      "required_token_positions_per_second": 28454.133156869626,
+      "required_pages_per_second_upper_estimate": 0.0017327074219542166
     },
     {
       "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": 28303.282475341784,
-      "required_pages_per_second_upper_estimate": 0.0021788650334522296
+      "required_token_positions_per_second": 28454.133156869626,
+      "required_pages_per_second_upper_estimate": 0.0021904779364976556
     },
     {
       "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": 28303.282475341784,
-      "required_pages_per_second_upper_estimate": 0.3514320433507802
+      "required_token_positions_per_second": 28454.133156869626,
+      "required_pages_per_second_upper_estimate": 0.3533051039506019
     }
   ],
   "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": 0.13052804964064535,
-    "producer_p95_seconds": 0.2851241200005461,
-    "gpu_update_seconds": 0.9768478276004674,
-    "producer_batches_per_second": 7.661188554897462,
-    "gpu_batches_per_second": 1.0237008997157764,
-    "producer_headroom": 7.483815396689151,
+    "producer_batch_seconds": 0.09065004438269852,
+    "producer_p95_seconds": 0.2790735769995081,
+    "gpu_update_seconds": 0.9716690312642681,
+    "producer_batches_per_second": 11.03143420182219,
+    "gpu_batches_per_second": 1.0291570152224256,
+    "producer_headroom": 10.718902984340081,
     "sustainable": true,
-    "buffer_seconds": 60.56456531122898,
+    "buffer_seconds": 60.24347993838462,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285102.8055065064,
-    "upload_bytes_per_second": 5252341.96801781,
+    "generation_bytes_per_second": 285672.97760275815,
+    "upload_bytes_per_second": 5179671.749066403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.176425208057346,
+    "estimated_full_shards_per_hour": 61.298770866985876,
     "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": 0.9499315409993869,
-      "bytes": 5779343,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.612262433000069,
       "bytes": 7066450,
@@ -562,6 +556,12 @@
       "bytes": 11337881,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.1246725650016742,
+      "bytes": 878351,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -185,3 +185,4 @@
 {"step": 213500, "loss": 0.5856230622157454, "lr": 0.0003, "grad_norm": 0.15907478251769827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31249621930, "tokens_per_second": 150298.50753169792, "data_wait_s": 1.1308002285659313e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213600, "loss": 0.6013360172510147, "lr": 0.0003, "grad_norm": 0.1541121413991341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31264166906, "tokens_per_second": 147387.321123808, "data_wait_s": 1.4970999473007396e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213700, "loss": 0.5764607787132263, "lr": 0.0003, "grad_norm": 0.145577945384662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31278567514, "tokens_per_second": 147023.58304407645, "data_wait_s": 1.2090000382158905e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 213800, "loss": 0.5735677741467953, "lr": 0.0003, "grad_norm": 0.1701035031124443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31293241918, "tokens_per_second": 149464.03457176004, "data_wait_s": 1.0456002200953662e-05, "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:ede25f6a675883a90338fda0015b36123c77f7dca936498f5fc466bbde89702a
-size 398580
+oid sha256:b6004e7931f1842a2aa870ad751a9c7ab01235c817f714c3292656294953cbb4
+size 400698