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": 75481.96784710741,
-      "required_pages_per_second_upper_estimate": 0.0065425856656887785
+      "required_token_positions_per_second": 75631.59871643831,
+      "required_pages_per_second_upper_estimate": 0.0065555552902064095
     },
     {
       "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": 9435.245980888427,
-      "required_pages_per_second_upper_estimate": 0.0037638606912751026
+      "required_token_positions_per_second": 9453.949839554789,
+      "required_pages_per_second_upper_estimate": 0.003771321940144722
     },
     {
       "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": 18870.491961776854,
-      "required_pages_per_second_upper_estimate": 0.004365356206041489
+      "required_token_positions_per_second": 18907.899679109578,
+      "required_pages_per_second_upper_estimate": 0.004374009823093075
     },
     {
       "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": 28305.73794266528,
-      "required_pages_per_second_upper_estimate": 0.0017236709319786898
+      "required_token_positions_per_second": 28361.84951866437,
+      "required_pages_per_second_upper_estimate": 0.0017270878325623472
     },
     {
       "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": 28305.73794266528,
-      "required_pages_per_second_upper_estimate": 0.0021790540621239606
+      "required_token_positions_per_second": 28361.84951866437,
+      "required_pages_per_second_upper_estimate": 0.0021833736865711543
     },
     {
       "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": 28305.73794266528,
-      "required_pages_per_second_upper_estimate": 0.35146253203701755
+      "required_token_positions_per_second": 28361.84951866437,
+      "required_pages_per_second_upper_estimate": 0.35215925001756176
     }
   ],
   "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": 0.21386824064063603,
-    "producer_p95_seconds": 0.1770654990004914,
-    "gpu_update_seconds": 0.9767630879647949,
-    "producer_batches_per_second": 4.675776061955386,
-    "gpu_batches_per_second": 1.0237897114679282,
-    "producer_headroom": 4.567125464907411,
+    "producer_batch_seconds": 0.07127279811719234,
+    "producer_p95_seconds": 0.1727572110003166,
+    "gpu_update_seconds": 0.9748306428960284,
+    "producer_batches_per_second": 14.030598298606453,
+    "gpu_batches_per_second": 1.025819210021136,
+    "producer_headroom": 13.677457159646451,
     "sustainable": true,
-    "buffer_seconds": 60.55931145381729,
+    "buffer_seconds": 60.43949985955376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411796.35923127295,
-    "upload_bytes_per_second": 3716072.9106241106,
+    "generation_bytes_per_second": 403917.07114514336,
+    "upload_bytes_per_second": 3749690.4025249705,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.36191256240502,
+    "estimated_full_shards_per_hour": 86.67120075956083,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 3672261,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0616448490000039,
+      "bytes": 5911843,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 3400, "loss": 0.8187024667859077, "lr": 0.0003, "grad_norm": 0.2072594781205535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 497011906, "tokens_per_second": 145624.0575270867, "data_wait_s": 1.6540000160603086e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3500, "loss": 0.8277444913983345, "lr": 0.0003, "grad_norm": 0.22791912088439592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 511687459, "tokens_per_second": 148971.41919640548, "data_wait_s": 1.034399974741973e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3600, "loss": 0.7926956005394459, "lr": 0.0003, "grad_norm": 0.19170133595960934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 526419490, "tokens_per_second": 150326.7971299855, "data_wait_s": 1.1434000043664128e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 3700, "loss": 0.7664323039352894, "lr": 0.0003, "grad_norm": 0.18385045648896226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 541090179, "tokens_per_second": 148339.78041008094, "data_wait_s": 1.9932999748561997e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f25a6001a12fe6b2f97ffab1de65e10fd77c7f9db35efd1d84278f5885dfdf2f
-size 75114
+oid sha256:81a977dda6e2310092ce65e01fcfcb064ac3a2e9cb3e208a7bd2969c83ae36b4
+size 77197