Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76087.57201776405,
-      "required_pages_per_second_upper_estimate": 0.005263914779679495
+      "required_token_positions_per_second": 74807.78120792736,
+      "required_pages_per_second_upper_estimate": 0.005175375881931194
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9510.946502220506,
-      "required_pages_per_second_upper_estimate": 0.003982034689331933
+      "required_token_positions_per_second": 9350.97265099092,
+      "required_pages_per_second_upper_estimate": 0.003915056978455995
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19021.89300444101,
-      "required_pages_per_second_upper_estimate": 0.003961758671772636
+      "required_token_positions_per_second": 18701.94530198184,
+      "required_pages_per_second_upper_estimate": 0.0038951220029387074
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28532.83950666152,
-      "required_pages_per_second_upper_estimate": 0.0017105001613010498
+      "required_token_positions_per_second": 28052.91795297276,
+      "required_pages_per_second_upper_estimate": 0.0016817295969551907
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28532.83950666152,
-      "required_pages_per_second_upper_estimate": 0.0022183027689454773
+      "required_token_positions_per_second": 28052.91795297276,
+      "required_pages_per_second_upper_estimate": 0.0021809909790979987
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28532.83950666152,
-      "required_pages_per_second_upper_estimate": 0.3659886290152964
+      "required_token_positions_per_second": 28052.91795297276,
+      "required_pages_per_second_upper_estimate": 0.35983271062419364
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3253223516094579,
-    "producer_p95_seconds": 0.32220884499838576,
-    "gpu_update_seconds": 0.9689887329140537,
-    "producer_batches_per_second": 3.0738742513470987,
-    "gpu_batches_per_second": 1.0320037437305236,
-    "producer_headroom": 2.9785495159499606,
+    "producer_batch_seconds": 0.10256466384365126,
+    "producer_p95_seconds": 0.3111375599983148,
+    "gpu_update_seconds": 0.9855659238852958,
+    "producer_batches_per_second": 9.749946643654892,
+    "gpu_batches_per_second": 1.014645469942591,
+    "producer_headroom": 9.609215171686074,
     "sustainable": true,
-    "buffer_seconds": 60.077301440671334,
+    "buffer_seconds": 60.119521357003045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448887.5120416302,
-    "upload_bytes_per_second": 5712213.984826128,
+    "generation_bytes_per_second": 450301.0805140784,
+    "upload_bytes_per_second": 4956521.627974124,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.3208105176609,
+    "estimated_full_shards_per_hour": 96.62412940565838,
     "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": 1.0634998090026784,
-      "bytes": 7147043,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 3.4036351569993712,
       "bytes": 6159372,
@@ -562,6 +556,12 @@
       "bytes": 11490913,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 13.443922794998798,
+      "bytes": 11916771,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -320,3 +320,4 @@
 {"step": 589100, "loss": 0.5802993774414062, "lr": 0.0003, "grad_norm": 0.1582367300546252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86226618796, "tokens_per_second": 153113.84424693335, "data_wait_s": 1.2295000487938523e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589200, "loss": 0.6048529744148254, "lr": 0.0003, "grad_norm": 0.16448242669712304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86241351148, "tokens_per_second": 153462.7825227095, "data_wait_s": 9.401999705005437e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 589300, "loss": 0.6123826503753662, "lr": 0.0003, "grad_norm": 0.167514450803692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86255947984, "tokens_per_second": 151317.76020781204, "data_wait_s": 1.6768997738836333e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 589400, "loss": 0.5678746700286865, "lr": 0.0003, "grad_norm": 0.1761332567785017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86270473152, "tokens_per_second": 150825.5584035463, "data_wait_s": 1.1059000826207921e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da60a7d3c5ac4b80135a17f0fa1813608b419da3356f35cfb46ebe62a88ef164
-size 686859
+oid sha256:e7002fbe979a052c67c43c43798f4b1b40a9717b33e11e4ce92608693b269da6
+size 688977