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": 75259.23593282516,
-      "required_pages_per_second_upper_estimate": 0.006523279827867668
+      "required_token_positions_per_second": 74658.5784203669,
+      "required_pages_per_second_upper_estimate": 0.006471216357040334
     },
     {
       "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": 9407.404491603145,
-      "required_pages_per_second_upper_estimate": 0.0037527543049318433
+      "required_token_positions_per_second": 9332.322302545863,
+      "required_pages_per_second_upper_estimate": 0.0037228028971381293
     },
     {
       "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": 18814.80898320629,
-      "required_pages_per_second_upper_estimate": 0.004352474928936142
+      "required_token_positions_per_second": 18664.644605091726,
+      "required_pages_per_second_upper_estimate": 0.004317737042862129
     },
     {
       "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": 28222.213474809436,
-      "required_pages_per_second_upper_estimate": 0.0017185847301052844
+      "required_token_positions_per_second": 27996.96690763759,
+      "required_pages_per_second_upper_estimate": 0.0017048683959419242
     },
     {
       "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": 28222.213474809436,
-      "required_pages_per_second_upper_estimate": 0.0021726241173778927
+      "required_token_positions_per_second": 27996.96690763759,
+      "required_pages_per_second_upper_estimate": 0.0021552840131145987
     },
     {
       "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": 28222.213474809436,
-      "required_pages_per_second_upper_estimate": 0.35042543768465967
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27996.96690763759,
+      "required_pages_per_second_upper_estimate": 0.3476286291721518
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.07976432144539558,
-    "producer_p95_seconds": 0.3128445969996392,
-    "gpu_update_seconds": 0.9796538469485405,
-    "producer_batches_per_second": 12.536933579815782,
-    "gpu_batches_per_second": 1.0207687165367996,
-    "producer_headroom": 12.281855210404869,
+    "producer_batch_seconds": 0.13128223560184438,
+    "producer_p95_seconds": 0.3594712480007729,
+    "gpu_update_seconds": 0.987535545947215,
+    "producer_batches_per_second": 7.61717680549577,
+    "gpu_batches_per_second": 1.012621777619994,
+    "producer_headroom": 7.522232855191729,
     "sustainable": true,
-    "buffer_seconds": 60.73853851080951,
+    "buffer_seconds": 60.23966830278012,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 546887.1852695104,
-    "upload_bytes_per_second": 5067628.434120764,
+    "generation_bytes_per_second": 540608.5542653536,
+    "upload_bytes_per_second": 4944737.065882058,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.34925907672866,
+    "estimated_full_shards_per_hour": 116.002011022286,
     "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": 1.1085641130011936,
-      "bytes": 12447049,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.0412157949976972,
       "bytes": 5733012,
@@ -562,6 +556,12 @@
       "bytes": 6431968,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9079114560008748,
+      "bytes": 1141999,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -287,3 +287,4 @@
 {"step": 383700, "loss": 0.5997693054378033, "lr": 0.0003, "grad_norm": 0.22822255922816126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56163022076, "tokens_per_second": 151180.54813497647, "data_wait_s": 1.0577998182270676e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383800, "loss": 0.6192993987351656, "lr": 0.0003, "grad_norm": 0.15225453381973653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56177615623, "tokens_per_second": 149306.57897811482, "data_wait_s": 1.5244000678649172e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383900, "loss": 0.6027638502418995, "lr": 0.0003, "grad_norm": 0.14847226563170499, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56192284609, "tokens_per_second": 149026.94777243093, "data_wait_s": 1.3251999916974455e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 384000, "loss": 0.6020997744053602, "lr": 0.0003, "grad_norm": 0.16682139343107447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56206755170, "tokens_per_second": 147282.2652653989, "data_wait_s": 1.4498997188638896e-05, "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:b915b502b0f7f9fa71dfd026b048511f41d60cbfc3cf3cb3cf090577db49a391
-size 616182
+oid sha256:daf01286181c1d3ba46e4f7f2bc0e0d2b347752272277a2ec2836f05718107bf
+size 618300