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": 75828.64753374553,
-      "required_pages_per_second_upper_estimate": 0.0065726349822749375
+      "required_token_positions_per_second": 75764.5799213499,
+      "required_pages_per_second_upper_estimate": 0.0065670817640103634
     },
     {
       "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": 9478.580941718192,
-      "required_pages_per_second_upper_estimate": 0.0037811476550654987
+      "required_token_positions_per_second": 9470.572490168737,
+      "required_pages_per_second_upper_estimate": 0.0037779529640054
     },
     {
       "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": 18957.161883436384,
-      "required_pages_per_second_upper_estimate": 0.004385405767078897
+      "required_token_positions_per_second": 18941.144980337474,
+      "required_pages_per_second_upper_estimate": 0.004381700538434831
     },
     {
       "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": 28435.742825154575,
-      "required_pages_per_second_upper_estimate": 0.0017315875472393835
+      "required_token_positions_per_second": 28411.717470506213,
+      "required_pages_per_second_upper_estimate": 0.0017301245291925925
     },
     {
       "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": 28435.742825154575,
-      "required_pages_per_second_upper_estimate": 0.0021890621978545347
+      "required_token_positions_per_second": 28411.717470506213,
+      "required_pages_per_second_upper_estimate": 0.0021872126595472653
     },
     {
       "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": 28435.742825154575,
-      "required_pages_per_second_upper_estimate": 0.35307675757918194
+      "required_token_positions_per_second": 28411.717470506213,
+      "required_pages_per_second_upper_estimate": 0.3527784430821388
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10271931223434194,
-    "producer_p95_seconds": 0.3838116410042858,
-    "gpu_update_seconds": 0.9722974416389176,
-    "producer_batches_per_second": 9.735267675065995,
-    "gpu_batches_per_second": 1.0284918556551856,
-    "producer_headroom": 9.46557585413672,
+    "producer_batch_seconds": 0.38371808113265615,
+    "producer_p95_seconds": 0.39793653800006723,
+    "gpu_update_seconds": 0.9731196302617391,
+    "producer_batches_per_second": 2.6060799560140806,
+    "gpu_batches_per_second": 1.0276228830478231,
+    "producer_headroom": 2.5360275632289513,
     "sustainable": true,
-    "buffer_seconds": 60.28244138161289,
+    "buffer_seconds": 60.333417076227825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478010.3710580445,
-    "upload_bytes_per_second": 5762842.910090436,
+    "generation_bytes_per_second": 422666.0849257531,
+    "upload_bytes_per_second": 5727061.897261182,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.56989811712266,
+    "estimated_full_shards_per_hour": 90.69430266217657,
     "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": 1.0899500639934558,
-      "bytes": 7493063,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8779368290051934,
       "bytes": 1248675,
@@ -562,6 +556,12 @@
       "bytes": 6335429,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0603171299953829,
+      "bytes": 4753623,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 395700, "loss": 0.6006899625062943, "lr": 0.0003, "grad_norm": 0.16791024958231635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57919014382, "tokens_per_second": 148759.88234629392, "data_wait_s": 1.3379001757130027e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395800, "loss": 0.6065213084220886, "lr": 0.0003, "grad_norm": 0.214611582458161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57933679130, "tokens_per_second": 150357.68617321388, "data_wait_s": 1.2053998943883926e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395900, "loss": 0.6089574210345745, "lr": 0.0003, "grad_norm": 0.15405435335908782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57948493160, "tokens_per_second": 150743.63286178105, "data_wait_s": 7.872004061937332e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 396000, "loss": 0.5834906883537769, "lr": 0.0003, "grad_norm": 0.15976082240941086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57963220400, "tokens_per_second": 150549.37824708482, "data_wait_s": 7.735994586255401e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3adce568cf600a405e5520d57b21f67580a01d3afeafa87cfa4f2ccd59b90ee
-size 872691
+oid sha256:9dfe51523fb610d86a0f12ce024b53ff782349c0c1d05efec277315d2e5103ff
+size 874809