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": 74299.59135141635,
-      "required_pages_per_second_upper_estimate": 0.006440100267748108
+      "required_token_positions_per_second": 74303.1604801283,
+      "required_pages_per_second_upper_estimate": 0.006440409630778986
     },
     {
       "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": 9287.448918927044,
-      "required_pages_per_second_upper_estimate": 0.0037049022334957092
+      "required_token_positions_per_second": 9287.895060016037,
+      "required_pages_per_second_upper_estimate": 0.003705080205846512
     },
     {
       "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": 18574.89783785409,
-      "required_pages_per_second_upper_estimate": 0.0042969757077508585
+      "required_token_positions_per_second": 18575.790120032074,
+      "required_pages_per_second_upper_estimate": 0.0042971821215291054
     },
     {
       "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": 27862.346756781135,
-      "required_pages_per_second_upper_estimate": 0.0016966707350521122
+      "required_token_positions_per_second": 27863.685180048113,
+      "required_pages_per_second_upper_estimate": 0.0016967522380069069
     },
     {
       "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": 27862.346756781135,
-      "required_pages_per_second_upper_estimate": 0.0021449205812492404
+      "required_token_positions_per_second": 27863.685180048113,
+      "required_pages_per_second_upper_estimate": 0.002145023616777325
     },
     {
       "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": 27862.346756781135,
-      "required_pages_per_second_upper_estimate": 0.3459570974431769
+      "required_token_positions_per_second": 27863.685180048113,
+      "required_pages_per_second_upper_estimate": 0.3459737161807382
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.10627422174994194,
+    "producer_p95_seconds": 0.3385915590006334,
+    "gpu_update_seconds": 0.992259273005189,
+    "producer_batches_per_second": 9.409619600442252,
+    "gpu_batches_per_second": 1.0078011132829903,
+    "producer_headroom": 9.336782303990207,
     "sustainable": true,
-    "buffer_seconds": 60.53072322737973,
+    "buffer_seconds": 60.52781565331653,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398908.3381198732,
+    "generation_bytes_per_second": 385591.8119941518,
     "upload_bytes_per_second": 2866688.096561006,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.59644324967525,
+    "estimated_full_shards_per_hour": 82.73902673595825,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -249,3 +249,4 @@
 {"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}
+{"step": 220200, "loss": 0.619529539719224, "lr": 0.0003, "grad_norm": 0.1527795924355659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32230628597, "tokens_per_second": 150216.34231026022, "data_wait_s": 9.638999472372234e-06, "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 sha256:76749f60aaefae6a9571648249dec743f9cb9c13f6f3e653d0aa5f488f140645
-size 535698
+oid sha256:7a3f072720228b59710ceac3fa1f0c670f5493684c0ea2251860ddb0d79a3558
+size 537816