Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77264.5377979312,
-      "required_pages_per_second_upper_estimate": 0.005345340003288319
+      "required_token_positions_per_second": 77397.00221962271,
+      "required_pages_per_second_upper_estimate": 0.005354504199340741
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9658.0672247414,
-      "required_pages_per_second_upper_estimate": 0.00404363106362139
+      "required_token_positions_per_second": 9674.625277452838,
+      "required_pages_per_second_upper_estimate": 0.00405056357452021
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19316.1344494828,
-      "required_pages_per_second_upper_estimate": 0.004023041405106119
+      "required_token_positions_per_second": 19349.250554905677,
+      "required_pages_per_second_upper_estimate": 0.004029938616535276
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28974.2016742242,
-      "required_pages_per_second_upper_estimate": 0.001736959149325424
+      "required_token_positions_per_second": 29023.875832358513,
+      "required_pages_per_second_upper_estimate": 0.0017399370392575282
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28974.2016742242,
-      "required_pages_per_second_upper_estimate": 0.0022526167361264686
+      "required_token_positions_per_second": 29023.875832358513,
+      "required_pages_per_second_upper_estimate": 0.002256478683427881
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28974.2016742242,
-      "required_pages_per_second_upper_estimate": 0.3716499490030169
+      "required_token_positions_per_second": 29023.875832358513,
+      "required_pages_per_second_upper_estimate": 0.3722871157676083
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09021152170321045,
-    "producer_p95_seconds": 0.22982535200026177,
-    "gpu_update_seconds": 0.9542281892997243,
-    "producer_batches_per_second": 11.085058550391485,
-    "gpu_batches_per_second": 1.047967363795725,
-    "producer_headroom": 10.577675348821494,
+    "producer_batch_seconds": 0.11206794107047813,
+    "producer_p95_seconds": 0.21562367700062168,
+    "gpu_update_seconds": 0.9525950345051931,
+    "producer_batches_per_second": 8.923158491607449,
+    "gpu_batches_per_second": 1.049764027501393,
+    "producer_headroom": 8.500156471208104,
     "sustainable": true,
-    "buffer_seconds": 60.11637592588263,
+    "buffer_seconds": 60.013487173827166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444918.8251620799,
+    "generation_bytes_per_second": 427813.7401465042,
     "upload_bytes_per_second": 4257258.899231691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.46922269961163,
+    "estimated_full_shards_per_hour": 91.79886964126915,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 700800, "loss": 0.48726826906204224, "lr": 8e-05, "grad_norm": 0.22323890264294283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102577989205, "tokens_per_second": 155513.06971923634, "data_wait_s": 1.8820999684976414e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700900, "loss": 0.5624342560768127, "lr": 8e-05, "grad_norm": 0.20474502207205159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102592700636, "tokens_per_second": 152826.1183855804, "data_wait_s": 1.684199924056884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701000, "loss": 0.5181252956390381, "lr": 8e-05, "grad_norm": 0.2046413997117008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102607411373, "tokens_per_second": 153609.74864793487, "data_wait_s": 1.1185000403202139e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 701100, "loss": 0.5583337545394897, "lr": 8e-05, "grad_norm": 0.20807786757317906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102621901511, "tokens_per_second": 134787.57889448781, "data_wait_s": 1.7554000805830583e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:969f103464da775bd981eab7f13ad23b317af5510d603309b6c542763a48f83c
-size 341394
+oid sha256:814e565c15ec4dc483ca06231f61bca57400e8b8b21d555764eb3fcf603c61b8
+size 343512