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": 74218.2968781337,
-      "required_pages_per_second_upper_estimate": 0.006433053868842927
+      "required_token_positions_per_second": 74299.59135141635,
+      "required_pages_per_second_upper_estimate": 0.006440100267748108
     },
     {
       "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": 9277.287109766712,
-      "required_pages_per_second_upper_estimate": 0.0037008485358890667
+      "required_token_positions_per_second": 9287.448918927044,
+      "required_pages_per_second_upper_estimate": 0.0037049022334957092
     },
     {
       "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": 18554.574219533424,
-      "required_pages_per_second_upper_estimate": 0.0042922741963358395
+      "required_token_positions_per_second": 18574.89783785409,
+      "required_pages_per_second_upper_estimate": 0.0042969757077508585
     },
     {
       "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": 27831.861329300136,
-      "required_pages_per_second_upper_estimate": 0.0016948143324631963
+      "required_token_positions_per_second": 27862.346756781135,
+      "required_pages_per_second_upper_estimate": 0.0016966707350521122
     },
     {
       "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": 27831.861329300136,
-      "required_pages_per_second_upper_estimate": 0.002142573728652689
+      "required_token_positions_per_second": 27862.346756781135,
+      "required_pages_per_second_upper_estimate": 0.0021449205812492404
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27831.861329300136,
-      "required_pages_per_second_upper_estimate": 0.34557857046202534
+      "required_token_positions_per_second": 27862.346756781135,
+      "required_pages_per_second_upper_estimate": 0.3459570974431769
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3495573840859265,
-    "producer_p95_seconds": 0.33007302099940716,
-    "gpu_update_seconds": 0.993393854362641,
-    "producer_batches_per_second": 2.8607606233664478,
-    "gpu_batches_per_second": 1.0066500770146172,
-    "producer_headroom": 2.8418620220548676,
+    "producer_batch_seconds": 0.134245814296861,
+    "producer_p95_seconds": 0.3406984260000172,
+    "gpu_update_seconds": 0.992306938153766,
+    "producer_batches_per_second": 7.4490218204396,
+    "gpu_batches_per_second": 1.0077527038766325,
+    "producer_headroom": 7.391716034881012,
     "sustainable": true,
-    "buffer_seconds": 60.5970251161211,
+    "buffer_seconds": 60.53072322737973,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384535.0366229114,
-    "upload_bytes_per_second": 2874108.1431070296,
+    "generation_bytes_per_second": 398908.3381198732,
+    "upload_bytes_per_second": 2866688.096561006,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.51226734176166,
+    "estimated_full_shards_per_hour": 85.59644324967525,
     "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": 1.1491692820018216,
-      "bytes": 11269048,
-      "pages": 45,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0205925500013109,
-      "bytes": 2657105,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0683662000010372,
       "bytes": 6645182,
@@ -562,6 +550,18 @@
       "bytes": 10747728,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.4399759389998508,
+      "bytes": 8283696,
+      "pages": 33,
+      "shards": 3
+    },
+    {
+      "seconds": 1.058994241997425,
+      "bytes": 5472606,
+      "pages": 16,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 219800, "loss": 0.5959744118154049, "lr": 0.0003, "grad_norm": 0.17277680729429762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32171710634, "tokens_per_second": 148777.3774970215, "data_wait_s": 2.0179999410174787e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219900, "loss": 0.5553730577230453, "lr": 0.0003, "grad_norm": 0.1439285963883817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32186398686, "tokens_per_second": 149145.27615156313, "data_wait_s": 1.767000139807351e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 220000, "loss": 0.6188193764537573, "lr": 0.0003, "grad_norm": 0.1620542855839367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32201121996, "tokens_per_second": 149783.29342078025, "data_wait_s": 1.0130999726243317e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 220100, "loss": 0.6336135417222977, "lr": 0.0003, "grad_norm": 0.15108254080626116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32215851555, "tokens_per_second": 140281.62330591143, "data_wait_s": 1.1740998161258176e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7e4540b84aa3fabf0e7aafbbdbb7daff6cb45459469d726594e5352432ada30
-size 533580
+oid sha256:76749f60aaefae6a9571648249dec743f9cb9c13f6f3e653d0aa5f488f140645
+size 535698