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": 74031.98176203418,
-      "required_pages_per_second_upper_estimate": 0.006416904546790765
+      "required_token_positions_per_second": 74748.756167329,
+      "required_pages_per_second_upper_estimate": 0.006479032735593599
     },
     {
       "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": 9253.997720254272,
-      "required_pages_per_second_upper_estimate": 0.003691558050205151
+      "required_token_positions_per_second": 9343.594520916125,
+      "required_pages_per_second_upper_estimate": 0.0037272995535807107
     },
     {
       "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": 18507.995440508545,
-      "required_pages_per_second_upper_estimate": 0.0042814990155669455
+      "required_token_positions_per_second": 18687.18904183225,
+      "required_pages_per_second_upper_estimate": 0.004322952301533514
     },
     {
       "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": 27761.993160762817,
-      "required_pages_per_second_upper_estimate": 0.0016905597275692237
+      "required_token_positions_per_second": 28030.783562748376,
+      "required_pages_per_second_upper_estimate": 0.0017069276528158938
     },
     {
       "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": 27761.993160762817,
-      "required_pages_per_second_upper_estimate": 0.0021371950836312167
+      "required_token_positions_per_second": 28030.783562748376,
+      "required_pages_per_second_upper_estimate": 0.00215788731283552
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27761.993160762817,
-      "required_pages_per_second_upper_estimate": 0.3447110416425099
+      "required_token_positions_per_second": 28030.783562748376,
+      "required_pages_per_second_upper_estimate": 0.34804851885156357
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24545751942176253,
-    "producer_p95_seconds": 0.33795177299907664,
-    "gpu_update_seconds": 0.9958939129441208,
-    "producer_batches_per_second": 4.074024712527666,
-    "gpu_batches_per_second": 1.0041230165206458,
-    "producer_headroom": 4.057296412390224,
+    "producer_batch_seconds": 0.07357237052346477,
+    "producer_p95_seconds": 0.3247366659998079,
+    "gpu_update_seconds": 0.9863441718676363,
+    "producer_batches_per_second": 13.592058987430145,
+    "gpu_batches_per_second": 1.0138448915924616,
+    "producer_headroom": 13.40644816593285,
     "sustainable": true,
-    "buffer_seconds": 60.749528689591365,
+    "buffer_seconds": 60.16699448392582,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 552693.1434522956,
-    "upload_bytes_per_second": 4009429.624898883,
+    "generation_bytes_per_second": 552739.715342531,
+    "upload_bytes_per_second": 4052136.1178426608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.59508254696513,
+    "estimated_full_shards_per_hour": 118.60507579047153,
     "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": 0.994495390999873,
-      "bytes": 2312576,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.5429994639998768,
       "bytes": 10370731,
@@ -562,6 +556,12 @@
       "bytes": 7079866,
       "pages": 44,
       "shards": 2
+    },
+    {
+      "seconds": 1.104166561999591,
+      "bytes": 7255454,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -205,3 +205,4 @@
 {"step": 135500, "loss": 0.5670278090983629, "lr": 0.0003, "grad_norm": 0.13061139796537177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19829721562, "tokens_per_second": 146977.373694967, "data_wait_s": 1.514999894425273e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 135600, "loss": 0.5967274568974972, "lr": 0.0003, "grad_norm": 0.13923674757594223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19844505097, "tokens_per_second": 149277.29960777762, "data_wait_s": 1.1913998605450615e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 135700, "loss": 0.5719640366733074, "lr": 0.0003, "grad_norm": 0.1361328808762725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19859178115, "tokens_per_second": 149631.42355862632, "data_wait_s": 1.4367000403581187e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 135800, "loss": 0.6057610101997852, "lr": 0.0003, "grad_norm": 0.1535781581093354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19873871396, "tokens_per_second": 150192.94375282497, "data_wait_s": 1.3393000699579716e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b8b1b6f3946262803f5d4c98f584b9339eb2880d5c7797863d6c2752e725e85
-size 441723
+oid sha256:ccdea84df526772abe04e1cb2096aa8d0381deb1eb7b1c5268ec13fd5b0a6e90
+size 443841