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": 76188.53281007441,
-      "required_pages_per_second_upper_estimate": 0.006603828925906231
+      "required_token_positions_per_second": 76028.11684352801,
+      "required_pages_per_second_upper_estimate": 0.00658992447649654
     },
     {
       "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": 9523.566601259301,
-      "required_pages_per_second_upper_estimate": 0.003799093107251995
+      "required_token_positions_per_second": 9503.514605441002,
+      "required_pages_per_second_upper_estimate": 0.003791094066316021
     },
     {
       "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": 19047.133202518602,
-      "required_pages_per_second_upper_estimate": 0.004406219048307434
+      "required_token_positions_per_second": 19007.029210882003,
+      "required_pages_per_second_upper_estimate": 0.0043969416956484535
     },
     {
       "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": 28570.6998037779,
-      "required_pages_per_second_upper_estimate": 0.0017398057191729994
+      "required_token_positions_per_second": 28510.543816323003,
+      "required_pages_per_second_upper_estimate": 0.0017361425351510704
     },
     {
       "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": 28570.6998037779,
-      "required_pages_per_second_upper_estimate": 0.0021994515596537854
+      "required_token_positions_per_second": 28510.543816323003,
+      "required_pages_per_second_upper_estimate": 0.0021948205852170763
     },
     {
       "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": 28570.6998037779,
-      "required_pages_per_second_upper_estimate": 0.3547524715817314
+      "required_token_positions_per_second": 28510.543816323003,
+      "required_pages_per_second_upper_estimate": 0.35400553554668046
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.25588165814821195,
-    "producer_p95_seconds": 0.3856714389985427,
-    "gpu_update_seconds": 0.9677046831153959,
-    "producer_batches_per_second": 3.908056588490525,
-    "gpu_batches_per_second": 1.0333731121158096,
-    "producer_headroom": 3.781844662562259,
+    "producer_batch_seconds": 0.11233758847657782,
+    "producer_p95_seconds": 0.38836248600273393,
+    "gpu_update_seconds": 0.9697464972299414,
+    "producer_batches_per_second": 8.901739956866692,
+    "gpu_batches_per_second": 1.031197331319553,
+    "producer_headroom": 8.632431142423284,
     "sustainable": true,
-    "buffer_seconds": 60.965395036269946,
+    "buffer_seconds": 60.12428282825637,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 542587.627276583,
-    "upload_bytes_per_second": 5683123.689065722,
+    "generation_bytes_per_second": 556436.6537091675,
+    "upload_bytes_per_second": 5586218.235565733,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.426674020034,
+    "estimated_full_shards_per_hour": 119.3983527036311,
     "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": 1.3039640329989197,
-      "bytes": 9738932,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1697420529999363,
       "bytes": 8456264,
@@ -562,6 +556,12 @@
       "bytes": 13274402,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.07868888899975,
+      "bytes": 1300513,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -312,3 +312,4 @@
 {"step": 311400, "loss": 0.5803543459624052, "lr": 0.0003, "grad_norm": 0.1543395265847266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45580506588, "tokens_per_second": 149135.81556980906, "data_wait_s": 1.5265999536495656e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 311500, "loss": 0.5991575866937637, "lr": 0.0003, "grad_norm": 0.15397654657214074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45595202813, "tokens_per_second": 150432.2112460396, "data_wait_s": 1.3062999641988426e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 311600, "loss": 0.5842957980930805, "lr": 0.0003, "grad_norm": 0.16250620971312138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45609751924, "tokens_per_second": 149430.65663408334, "data_wait_s": 1.084099494619295e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 311700, "loss": 0.5714015923440456, "lr": 0.0003, "grad_norm": 0.14300842317176646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45624203878, "tokens_per_second": 148448.59188599358, "data_wait_s": 9.22300387173891e-06, "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 sha256:99366b39c1ddaa1e8e72fe31f20d94fc909c5e1ce7cd0bd7cbb92a3d4be6cc94
-size 669915
+oid sha256:1d575404fcea129a8e31fac6d3b54862a674182c6bb5dc52bde69021a18d8c79
+size 672033