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": 76168.4800095302,
-      "required_pages_per_second_upper_estimate": 0.006602090799978417
+      "required_token_positions_per_second": 76223.21788369953,
+      "required_pages_per_second_upper_estimate": 0.006606835340179541
     },
     {
       "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": 9521.060001191276,
-      "required_pages_per_second_upper_estimate": 0.003798093187007849
+      "required_token_positions_per_second": 9527.902235462441,
+      "required_pages_per_second_upper_estimate": 0.0038008226565591358
     },
     {
       "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": 19042.12000238255,
-      "required_pages_per_second_upper_estimate": 0.004405059332685268
+      "required_token_positions_per_second": 19055.804470924882,
+      "required_pages_per_second_upper_estimate": 0.004408224993644115
     },
     {
       "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": 28563.180003573827,
-      "required_pages_per_second_upper_estimate": 0.0017393478027939138
+      "required_token_positions_per_second": 28583.706706387322,
+      "required_pages_per_second_upper_estimate": 0.0017405977712999692
     },
     {
       "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": 28563.180003573827,
-      "required_pages_per_second_upper_estimate": 0.002198872664617937
+      "required_token_positions_per_second": 28583.706706387322,
+      "required_pages_per_second_upper_estimate": 0.0022004528670220695
     },
     {
       "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": 28563.180003573827,
-      "required_pages_per_second_upper_estimate": 0.3546591008303491
+      "required_token_positions_per_second": 28583.706706387322,
+      "required_pages_per_second_upper_estimate": 0.35491397378083767
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08096750484358495,
-    "producer_p95_seconds": 0.21001365999836707,
-    "gpu_update_seconds": 0.967959449772073,
-    "producer_batches_per_second": 12.350633775016595,
-    "gpu_batches_per_second": 1.033101128601484,
-    "producer_headroom": 11.954912673201445,
+    "producer_batch_seconds": 0.2671125667264107,
+    "producer_p95_seconds": 0.22508589299832238,
+    "gpu_update_seconds": 0.9672643329292828,
+    "producer_batches_per_second": 3.7437399979172348,
+    "gpu_batches_per_second": 1.033843558535421,
+    "producer_headroom": 3.6211861717460887,
     "sustainable": true,
-    "buffer_seconds": 60.013485885868526,
+    "buffer_seconds": 60.93765297454482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444621.45791565435,
-    "upload_bytes_per_second": 3885106.2770338436,
+    "generation_bytes_per_second": 416977.023595233,
+    "upload_bytes_per_second": 3945714.5081224507,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.40541461088392,
+    "estimated_full_shards_per_hour": 89.47356253521673,
     "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": 1.4124847809998755,
-      "bytes": 4874448,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2419382850002876,
       "bytes": 7186580,
@@ -562,6 +556,12 @@
       "bytes": 1167322,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.291220240000257,
+      "bytes": 10903036,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -115,3 +115,4 @@
 {"step": 291700, "loss": 0.5707413833588362, "lr": 0.0003, "grad_norm": 0.14450361581721619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42696551761, "tokens_per_second": 149938.74250766585, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 291800, "loss": 0.5925162993371487, "lr": 0.0003, "grad_norm": 0.15542837092778156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42711194406, "tokens_per_second": 150613.9168806891, "data_wait_s": 9.371997293783352e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 291900, "loss": 0.6618529204279184, "lr": 0.0003, "grad_norm": 0.14905779807584116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42725726528, "tokens_per_second": 149057.4284896329, "data_wait_s": 1.3377000868786126e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 292000, "loss": 0.6027472764253616, "lr": 0.0003, "grad_norm": 0.1884136091430959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42740315641, "tokens_per_second": 149508.34723971388, "data_wait_s": 1.237399919773452e-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 sha256:6d938058c0c15741f4390df807e358f61c7ddeae5126feeccb405ec651f1c624
-size 249537
+oid sha256:ddccf7e6da04e612472baca55ec3c67a91e878482d597c0570c8aa93342e0d8b
+size 251655