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": 76126.03867492927,
-      "required_pages_per_second_upper_estimate": 0.006598412092661788
+      "required_token_positions_per_second": 75848.3081906056,
+      "required_pages_per_second_upper_estimate": 0.006574339118182089
     },
     {
       "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": 9515.754834366158,
-      "required_pages_per_second_upper_estimate": 0.0037959768766419972
+      "required_token_positions_per_second": 9481.0385238257,
+      "required_pages_per_second_upper_estimate": 0.0037821280213123105
     },
     {
       "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": 19031.509668732317,
-      "required_pages_per_second_upper_estimate": 0.00440260481873078
+      "required_token_positions_per_second": 18962.0770476514,
+      "required_pages_per_second_upper_estimate": 0.004386542803816106
     },
     {
       "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": 28547.264503098475,
-      "required_pages_per_second_upper_estimate": 0.0017383786323171425
+      "required_token_positions_per_second": 28443.1155714771,
+      "required_pages_per_second_upper_estimate": 0.001732036508808615
     },
     {
       "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": 28547.264503098475,
-      "required_pages_per_second_upper_estimate": 0.0021976474453344187
+      "required_token_positions_per_second": 28443.1155714771,
+      "required_pages_per_second_upper_estimate": 0.0021896297722755107
     },
     {
       "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": 28547.264503098475,
-      "required_pages_per_second_upper_estimate": 0.35446148358019886
+      "required_token_positions_per_second": 28443.1155714771,
+      "required_pages_per_second_upper_estimate": 0.35316830241351305
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12028842567906395,
-    "producer_p95_seconds": 0.37660223899729317,
-    "gpu_update_seconds": 0.9684991007456819,
-    "producer_batches_per_second": 8.313351798850992,
-    "gpu_batches_per_second": 1.0325254811595224,
-    "producer_headroom": 8.051473741369682,
+    "producer_batch_seconds": 0.145713389890318,
+    "producer_p95_seconds": 0.3763052099966444,
+    "gpu_update_seconds": 0.9720454122025068,
+    "producer_batches_per_second": 6.86278728916213,
+    "gpu_batches_per_second": 1.0287585203803928,
+    "producer_headroom": 6.670940899351726,
     "sustainable": true,
-    "buffer_seconds": 60.04694424623228,
+    "buffer_seconds": 60.26681555655542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407302.24403508316,
+    "generation_bytes_per_second": 440714.27203669265,
     "upload_bytes_per_second": 5508692.260353463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.39758005895015,
+    "estimated_full_shards_per_hour": 94.56702347589096,
     "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": [

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

@@ -354,3 +354,4 @@
 {"step": 75400, "loss": 0.6059148088097572, "lr": 0.0003, "grad_norm": 0.13393706547614445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11034406397, "tokens_per_second": 150402.01004284364, "data_wait_s": 8.873001206666231e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 75500, "loss": 0.6612876560539007, "lr": 0.0003, "grad_norm": 0.1428170348202804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11049113914, "tokens_per_second": 151177.2646087947, "data_wait_s": 1.1476004146970809e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 75600, "loss": 0.6527904272079468, "lr": 0.0003, "grad_norm": 0.1283468365167278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11063941581, "tokens_per_second": 151668.4325913146, "data_wait_s": 9.27999644773081e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 75700, "loss": 0.6094121038913727, "lr": 0.0003, "grad_norm": 0.1716657095901446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11078685161, "tokens_per_second": 150022.6171831484, "data_wait_s": 1.651100319577381e-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 sha256:465b4b0a588209713772f6b368fe327d494033a0e9c922ae100d3754415d5e88
-size 759654
+oid sha256:eb4b0d29f6f8ef85d8c8ebe298c8e8229081629dbf65dfe4fcf28f7ea13bc1bb
+size 761772