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": 76229.27411781656,
-      "required_pages_per_second_upper_estimate": 0.00660736027920342
+      "required_token_positions_per_second": 75271.7260066796,
+      "required_pages_per_second_upper_estimate": 0.006524362435813569
     },
     {
       "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": 9528.65926472707,
-      "required_pages_per_second_upper_estimate": 0.0038011246468513922
+      "required_token_positions_per_second": 9408.96575083495,
+      "required_pages_per_second_upper_estimate": 0.0037533771145823163
     },
     {
       "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": 19057.31852945414,
-      "required_pages_per_second_upper_estimate": 0.004408575244438339
+      "required_token_positions_per_second": 18817.9315016699,
+      "required_pages_per_second_upper_estimate": 0.0043531972686282495
     },
     {
       "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": 28585.97779418121,
-      "required_pages_per_second_upper_estimate": 0.0017407360686311402
+      "required_token_positions_per_second": 28226.897252504852,
+      "required_pages_per_second_upper_estimate": 0.001718869947593053
     },
     {
       "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": 28585.97779418121,
-      "required_pages_per_second_upper_estimate": 0.0022006277016472166
+      "required_token_positions_per_second": 28226.897252504852,
+      "required_pages_per_second_upper_estimate": 0.002172984687550412
     },
     {
       "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": 28585.97779418121,
-      "required_pages_per_second_upper_estimate": 0.35494217309039583
+      "required_token_positions_per_second": 28226.897252504852,
+      "required_pages_per_second_upper_estimate": 0.3504835945280412
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 0.054874912054685865,
-    "producer_p95_seconds": 0.1411703599997054,
-    "gpu_update_seconds": 0.9671874860837492,
-    "producer_batches_per_second": 18.22326382962482,
-    "gpu_batches_per_second": 1.033925701467781,
-    "producer_headroom": 17.625312731615747,
+    "producer_batch_seconds": 0.08681666446094383,
+    "producer_p95_seconds": 0.17449106000003667,
+    "gpu_update_seconds": 0.979491289909539,
+    "producer_batches_per_second": 11.518525921366969,
+    "gpu_batches_per_second": 1.0209381240055284,
+    "producer_headroom": 11.282295812576194,
     "sustainable": true,
-    "buffer_seconds": 60.932811623276194,
+    "buffer_seconds": 60.72845997439142,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 323343.7827829363,
+    "upload_bytes_per_second": 42894.73603399726,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 69.38204872718875,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.0642098359999181,
+      "bytes": 45649,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 157800, "loss": 0.602766864001751, "lr": 0.0003, "grad_norm": 0.13362723631035256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23094887446, "tokens_per_second": 148526.46023734435, "data_wait_s": 1.6574999790464062e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157900, "loss": 0.6427027005702257, "lr": 0.0003, "grad_norm": 0.16760281105504934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23109366651, "tokens_per_second": 149030.92749643698, "data_wait_s": 1.3122999916959088e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 158000, "loss": 0.5854500038549304, "lr": 0.0003, "grad_norm": 0.13975316095131562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23124097242, "tokens_per_second": 149803.65281622653, "data_wait_s": 1.4758000361325685e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 158100, "loss": 0.5768682304769754, "lr": 0.0003, "grad_norm": 0.14110618002533135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23138787938, "tokens_per_second": 149399.6597650285, "data_wait_s": 1.816199983295519e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84ea54f5cc68e137a28cd9973c3dc8d42e92deb894145a5dfd6e3d874b2795db
-size 63705
+oid sha256:69e33a89413d69b9d7066a6e84ef9bdd1ad345e6cb452575e3c5741f73ece959
+size 65823