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": 76192.70134475944,
-      "required_pages_per_second_upper_estimate": 0.006604190243927674
+      "required_token_positions_per_second": 75621.19981057389,
+      "required_pages_per_second_upper_estimate": 0.006554653939401868
     },
     {
       "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": 9524.08766809493,
-      "required_pages_per_second_upper_estimate": 0.0037993009686033705
+      "required_token_positions_per_second": 9452.649976321736,
+      "required_pages_per_second_upper_estimate": 0.00377080340526637
     },
     {
       "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": 19048.17533618986,
-      "required_pages_per_second_upper_estimate": 0.004406460127591353
+      "required_token_positions_per_second": 18905.299952643472,
+      "required_pages_per_second_upper_estimate": 0.004373408422128766
     },
     {
       "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": 28572.26300428479,
-      "required_pages_per_second_upper_estimate": 0.0017399009098823897
+      "required_token_positions_per_second": 28357.94992896521,
+      "required_pages_per_second_upper_estimate": 0.001726850368009234
     },
     {
       "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": 28572.26300428479,
-      "required_pages_per_second_upper_estimate": 0.0021995718991560218
+      "required_token_positions_per_second": 28357.94992896521,
+      "required_pages_per_second_upper_estimate": 0.0021830734853613577
     },
     {
       "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": 28572.26300428479,
-      "required_pages_per_second_upper_estimate": 0.3547718813003314
+      "required_token_positions_per_second": 28357.94992896521,
+      "required_pages_per_second_upper_estimate": 0.35211083016458533
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07261075600004574,
-    "producer_p95_seconds": 0.1747196010001062,
-    "gpu_update_seconds": 0.967651739585829,
-    "producer_batches_per_second": 13.772064293055564,
-    "gpu_batches_per_second": 1.033429651485995,
-    "producer_headroom": 13.326561970863096,
+    "producer_batch_seconds": 0.24531029464067444,
+    "producer_p95_seconds": 0.19242204999954993,
+    "gpu_update_seconds": 0.9749646948829663,
+    "producer_batches_per_second": 4.076469768481505,
+    "gpu_batches_per_second": 1.0256781658335217,
+    "producer_headroom": 3.974414104027207,
     "sustainable": true,
-    "buffer_seconds": 60.96205959390723,
+    "buffer_seconds": 60.44781108274391,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416889.0103106129,
-    "upload_bytes_per_second": 5977966.140172593,
+    "generation_bytes_per_second": 380799.95822302793,
+    "upload_bytes_per_second": 6135503.703642934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.45467693318167,
+    "estimated_full_shards_per_hour": 81.71080646532181,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 2586371,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9901489570002013,
+      "bytes": 9709695,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 45100, "loss": 0.6282200925052166, "lr": 0.0003, "grad_norm": 0.13956308911328189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6599624209, "tokens_per_second": 138713.12959167766, "data_wait_s": 1.050100036081858e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45200, "loss": 0.6508970856666565, "lr": 0.0003, "grad_norm": 0.12931241013622158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6614419603, "tokens_per_second": 151655.39440942844, "data_wait_s": 1.0557000678090844e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45300, "loss": 0.6566140204668045, "lr": 0.0003, "grad_norm": 0.13486927166703083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6629087250, "tokens_per_second": 148765.899250805, "data_wait_s": 1.499800055171363e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 45400, "loss": 0.6239417102187872, "lr": 0.0003, "grad_norm": 0.1296677709303981, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6643543499, "tokens_per_second": 146850.14593881185, "data_wait_s": 1.4342000213218853e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0fd03a06a4475bc05ec09afe1d9d90106e65303e51a4fcf38407742849c18915
-size 113202
+oid sha256:16bdeb0416e04518d51e73ea8e7b7ebaad389801188d34aa8fd4d07aed522f14
+size 115320