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": 76272.00656720882,
-      "required_pages_per_second_upper_estimate": 0.0066110642196123386
+      "required_token_positions_per_second": 75879.19152795528,
+      "required_pages_per_second_upper_estimate": 0.006577016007590457
     },
     {
       "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": 9534.000820901103,
-      "required_pages_per_second_upper_estimate": 0.003803255473472596
+      "required_token_positions_per_second": 9484.89894099441,
+      "required_pages_per_second_upper_estimate": 0.003783667999439289
     },
     {
       "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": 19068.001641802206,
-      "required_pages_per_second_upper_estimate": 0.004411046594463708
+      "required_token_positions_per_second": 18969.79788198882,
+      "required_pages_per_second_upper_estimate": 0.004388328882958024
     },
     {
       "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": 28602.002462703313,
-      "required_pages_per_second_upper_estimate": 0.001741711886869198
+      "required_token_positions_per_second": 28454.69682298323,
+      "required_pages_per_second_upper_estimate": 0.0017327417462632076
     },
     {
       "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": 28602.002462703313,
-      "required_pages_per_second_upper_estimate": 0.0022018613249891695
+      "required_token_positions_per_second": 28454.69682298323,
+      "required_pages_per_second_upper_estimate": 0.0021905213290754084
     },
     {
       "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": 28602.002462703313,
-      "required_pages_per_second_upper_estimate": 0.3551411458423248
+      "required_token_positions_per_second": 28454.69682298323,
+      "required_pages_per_second_upper_estimate": 0.35331210279726377
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1327567917340957,
-    "producer_p95_seconds": 0.37840703500114614,
-    "gpu_update_seconds": 0.9666456058820595,
-    "producer_batches_per_second": 7.53257130530047,
-    "gpu_batches_per_second": 1.0345052974068039,
-    "producer_headroom": 7.281326953261988,
+    "producer_batch_seconds": 0.19016914689063924,
+    "producer_p95_seconds": 0.3627977700016345,
+    "gpu_update_seconds": 0.9716497832325646,
+    "producer_batches_per_second": 5.2584765528504525,
+    "gpu_batches_per_second": 1.0291774024516505,
+    "producer_headroom": 5.109397602710666,
     "sustainable": true,
-    "buffer_seconds": 60.89867317056975,
+    "buffer_seconds": 60.242286560419004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299442.34208720096,
+    "generation_bytes_per_second": 302683.31541065447,
     "upload_bytes_per_second": 5687702.698158323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.25335595094701,
+    "estimated_full_shards_per_hour": 64.94879338016248,
     "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": [

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -289,3 +289,4 @@
 {"step": 309100, "loss": 0.5872989073395729, "lr": 0.0003, "grad_norm": 0.15178216332064237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45243936418, "tokens_per_second": 148635.3064319802, "data_wait_s": 1.1945994629058987e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309200, "loss": 0.5954781509935856, "lr": 0.0003, "grad_norm": 0.13767809501678832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45258524838, "tokens_per_second": 149608.27499610386, "data_wait_s": 1.2308002624195069e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 309300, "loss": 0.6228629481047392, "lr": 0.0003, "grad_norm": 0.15283185855990575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45273186809, "tokens_per_second": 123683.6008653607, "data_wait_s": 1.4411998563446105e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 309400, "loss": 0.5916283000260592, "lr": 0.0003, "grad_norm": 0.16134914729114636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45287881087, "tokens_per_second": 150733.95974090102, "data_wait_s": 1.7038000805769116e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:417b15694a2ed85006500abdbfb6520bd58fd0b1dd9da9420cac698910d592b5
-size 620418
+oid sha256:83d3b38c06e39829e12116c58bed25eea7917f9602adc5d0401a891269815d2b
+size 622536