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": 75967.67214178745,
-      "required_pages_per_second_upper_estimate": 0.006584685282945364
+      "required_token_positions_per_second": 74944.48117012672,
+      "required_pages_per_second_upper_estimate": 0.006495997682775621
     },
     {
       "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": 9495.959017723431,
-      "required_pages_per_second_upper_estimate": 0.003788080029409379
+      "required_token_positions_per_second": 9368.06014626584,
+      "required_pages_per_second_upper_estimate": 0.003737059257326408
     },
     {
       "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": 18991.918035446863,
-      "required_pages_per_second_upper_estimate": 0.0043934459911591376
+      "required_token_positions_per_second": 18736.12029253168,
+      "required_pages_per_second_upper_estimate": 0.004334271685222227
     },
     {
       "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": 28487.877053170298,
-      "required_pages_per_second_upper_estimate": 0.001734762248198386
+      "required_token_positions_per_second": 28104.18043879752,
+      "required_pages_per_second_upper_estimate": 0.0017113971374836388
     },
     {
       "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": 28487.877053170298,
-      "required_pages_per_second_upper_estimate": 0.0021930756350439645
+      "required_token_positions_per_second": 28104.18043879752,
+      "required_pages_per_second_upper_estimate": 0.00216353760752934
     },
     {
       "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": 28487.877053170298,
-      "required_pages_per_second_upper_estimate": 0.3537240902091001
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28104.18043879752,
+      "required_pages_per_second_upper_estimate": 0.3489598624085516
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1531781848828091,
-    "producer_p95_seconds": 0.13073813200003315,
-    "gpu_update_seconds": 0.9705180890944336,
-    "producer_batches_per_second": 6.5283447559132695,
-    "gpu_batches_per_second": 1.0303774975828377,
-    "producer_headroom": 6.335876677458613,
+    "producer_batch_seconds": 0.15572974917968363,
+    "producer_p95_seconds": 0.1324537320000445,
+    "gpu_update_seconds": 0.9837682354840076,
+    "producer_batches_per_second": 6.421380662767157,
+    "gpu_batches_per_second": 1.0164995818430815,
+    "producer_headroom": 6.317150323981573,
     "sustainable": true,
-    "buffer_seconds": 60.172121523854884,
+    "buffer_seconds": 60.009862364524466,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 98,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 687225.4147494808,
-    "upload_bytes_per_second": 1764455.250305986,
+    "generation_bytes_per_second": 637032.4643789999,
+    "upload_bytes_per_second": 3501156.05271489,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.73128810817393,
+    "estimated_full_shards_per_hour": 68.34616874946354,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 775414,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.4328443340000376,
+      "bytes": 11483713,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -2,3 +2,4 @@
 {"step": 237300, "loss": 0.5835124477744102, "lr": 0.0003, "grad_norm": 0.15055686807409902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34733571560, "tokens_per_second": 149880.6769248668, "data_wait_s": 8.639999919068941e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 237400, "loss": 0.6066939048469067, "lr": 0.0003, "grad_norm": 0.18699914909229576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34748361812, "tokens_per_second": 151019.1976334992, "data_wait_s": 1.131900000927999e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 237500, "loss": 0.5953725799918175, "lr": 0.0003, "grad_norm": 0.16524226536258613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34762871164, "tokens_per_second": 147611.56265858695, "data_wait_s": 8.885000056579884e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 237600, "loss": 0.5809929296374321, "lr": 0.0003, "grad_norm": 0.1370142476338837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34777355128, "tokens_per_second": 148249.4230582852, "data_wait_s": 1.2894000064989086e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f38820bf7695d2ae546d6a497bd3339a4e8198878bbaef1bdc62ddd4def4eb4
-size 8637
+oid sha256:7426a96bde95f9ab20cc8d4711f587d5134db34c0d1ba597340911c43a4fb4b9
+size 10755