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": 77223.44987889791,
-      "required_pages_per_second_upper_estimate": 0.0053424974457124885
+      "required_token_positions_per_second": 74860.70388277741,
+      "required_pages_per_second_upper_estimate": 0.005179037195374844
     },
     {
       "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": 9652.931234862239,
-      "required_pages_per_second_upper_estimate": 0.00404148073191065
+      "required_token_positions_per_second": 9357.587985347176,
+      "required_pages_per_second_upper_estimate": 0.00391782668080707
     },
     {
       "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": 19305.862469724478,
-      "required_pages_per_second_upper_estimate": 0.004020902022612784
+      "required_token_positions_per_second": 18715.175970694352,
+      "required_pages_per_second_upper_estimate": 0.003897877602315326
     },
     {
       "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": 28958.79370458672,
-      "required_pages_per_second_upper_estimate": 0.0017360354650722643
+      "required_token_positions_per_second": 28072.76395604153,
+      "required_pages_per_second_upper_estimate": 0.0016829193345360713
     },
     {
       "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": 28958.79370458672,
-      "required_pages_per_second_upper_estimate": 0.0022514188342596503
+      "required_token_positions_per_second": 28072.76395604153,
+      "required_pages_per_second_upper_estimate": 0.0021825339185432536
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28958.79370458672,
-      "required_pages_per_second_upper_estimate": 0.3714523121122961
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28072.76395604153,
+      "required_pages_per_second_upper_estimate": 0.36008727384258193
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1938344892498094,
-    "producer_p95_seconds": 0.24779992799813044,
-    "gpu_update_seconds": 0.9547359010199687,
-    "producer_batches_per_second": 5.159040601444375,
-    "gpu_batches_per_second": 1.0474100732272396,
-    "producer_headroom": 4.925521277018597,
-    "sustainable": true,
-    "buffer_seconds": 60.148361764258034,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 18.944940262664062,
+    "producer_p95_seconds": 0.280350291999639,
+    "gpu_update_seconds": 0.9848691793687769,
+    "producer_batches_per_second": 0.05278454226486849,
+    "gpu_batches_per_second": 1.015363279660067,
+    "producer_headroom": 0.05198586882375755,
+    "sustainable": false,
+    "buffer_seconds": 60.077019941495394,
+    "buffer_drain_seconds": 63.371439270588944,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302461.01880952006,
-    "upload_bytes_per_second": 5352014.540450696,
+    "generation_bytes_per_second": 38138.65869711697,
+    "upload_bytes_per_second": 5263667.821253658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.90109370435907,
+    "estimated_full_shards_per_hour": 8.18366833386547,
     "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": 0.9681137329971534,
-      "bytes": 5823551,
-      "pages": 34,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3083030150010018,
-      "bytes": 7346473,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8391288720013108,
       "bytes": 1133832,
@@ -562,6 +550,18 @@
       "bytes": 5165514,
       "pages": 4,
       "shards": 2
+    },
+    {
+      "seconds": 1.7178185550001217,
+      "bytes": 4032890,
+      "pages": 3,
+      "shards": 2
+    },
+    {
+      "seconds": 1.248450455001148,
+      "bytes": 5769233,
+      "pages": 1,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -249,3 +249,4 @@
 {"step": 709900, "loss": 0.6061674356460571, "lr": 8e-05, "grad_norm": 0.2592011926502791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103909177369, "tokens_per_second": 152417.59228703275, "data_wait_s": 1.621200135559775e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710000, "loss": 0.5671007633209229, "lr": 8e-05, "grad_norm": 0.22037535991920845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103923858867, "tokens_per_second": 152737.25572226194, "data_wait_s": 1.7985999875236303e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 710100, "loss": 0.5553442239761353, "lr": 8e-05, "grad_norm": 0.21565878596663604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103938573748, "tokens_per_second": 6114.882245130367, "data_wait_s": 1.7717000446282327e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 710200, "loss": 0.5633423924446106, "lr": 8e-05, "grad_norm": 0.2694930185439089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103953324137, "tokens_per_second": 153774.64411334082, "data_wait_s": 1.759799852152355e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6138795f0067ea31379c3939f5b188c2f09f3b08aaf985442e2389bd508f1ebf
-size 535698
+oid sha256:8e9200fe37b4c83f48f03165463f65e4f28b5c0dee4ffc9dcf1205a61266dd9d
+size 537816