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": 75050.54241632819,
-      "required_pages_per_second_upper_estimate": 0.0065051908027865925
+      "required_token_positions_per_second": 73902.64013814388,
+      "required_pages_per_second_upper_estimate": 0.0064056935426453025
     },
     {
       "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": 9381.317802041023,
-      "required_pages_per_second_upper_estimate": 0.0037423479344347466
+      "required_token_positions_per_second": 9237.830017267985,
+      "required_pages_per_second_upper_estimate": 0.0036851085117552197
     },
     {
       "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": 18762.635604082046,
-      "required_pages_per_second_upper_estimate": 0.004340405535895859
+      "required_token_positions_per_second": 18475.66003453597,
+      "required_pages_per_second_upper_estimate": 0.0042740187884788994
     },
     {
       "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": 28143.953406123066,
-      "required_pages_per_second_upper_estimate": 0.001713819102521132
+      "required_token_positions_per_second": 27713.490051803954,
+      "required_pages_per_second_upper_estimate": 0.0016876061427097747
     },
     {
       "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": 28143.953406123066,
-      "required_pages_per_second_upper_estimate": 0.0021665994406526805
+      "required_token_positions_per_second": 27713.490051803954,
+      "required_pages_per_second_upper_estimate": 0.0021334611800383645
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28143.953406123066,
-      "required_pages_per_second_upper_estimate": 0.34945370955117605
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27713.490051803954,
+      "required_pages_per_second_upper_estimate": 0.34410879535870414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07359510257026614,
-    "producer_p95_seconds": 0.27992867099965224,
-    "gpu_update_seconds": 0.9823779765775491,
-    "producer_batches_per_second": 13.587860673816353,
-    "gpu_batches_per_second": 1.017938129561743,
-    "producer_headroom": 13.34841507476136,
+    "producer_batch_seconds": 0.180633329914059,
+    "producer_p95_seconds": 0.31270402899826877,
+    "gpu_update_seconds": 0.9976368890500065,
+    "producer_batches_per_second": 5.536076871725589,
+    "gpu_batches_per_second": 1.0023687084709185,
+    "producer_headroom": 5.522994507850008,
     "sustainable": true,
-    "buffer_seconds": 60.907434547808045,
+    "buffer_seconds": 60.85585023205039,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294756.2643051768,
-    "upload_bytes_per_second": 5276812.650486299,
+    "generation_bytes_per_second": 334798.43467378593,
+    "upload_bytes_per_second": 5277827.151625277,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.24783274523237,
+    "estimated_full_shards_per_hour": 71.8399503723162,
     "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": 0.9227722110008472,
-      "bytes": 1091891,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.4008473989997583,
       "bytes": 11311428,
@@ -562,6 +556,12 @@
       "bytes": 6203437,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.8579733459991985,
+      "bytes": 830240,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -203,3 +203,4 @@
 {"step": 175300, "loss": 0.6606020256876945, "lr": 0.0003, "grad_norm": 0.16128645448699025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25656247799, "tokens_per_second": 149726.67388678485, "data_wait_s": 1.64319972100202e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 175400, "loss": 0.6293471790850163, "lr": 0.0003, "grad_norm": 0.1479959484642054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25670912341, "tokens_per_second": 149688.63152222676, "data_wait_s": 9.431001672055572e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 175500, "loss": 0.6122764702886343, "lr": 0.0003, "grad_norm": 0.14353583252303045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25685544053, "tokens_per_second": 149100.9483063834, "data_wait_s": 1.197900201077573e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 175600, "loss": 0.6353216134011745, "lr": 0.0003, "grad_norm": 0.1545032076960217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25700041187, "tokens_per_second": 149003.36608266132, "data_wait_s": 9.59899989538826e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:28b147b892d188d607d42e5347513046e0166793f6aba8f72afa2ca56c29df26
-size 437487
+oid sha256:8c7a12de0b6201ab8d0b060b74ef9999117efe36bcb07b207fbd811ff1fd97e0
+size 439605