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": 77489.84512951208,
-      "required_pages_per_second_upper_estimate": 0.005360927287272119
+      "required_token_positions_per_second": 77448.86840404927,
+      "required_pages_per_second_upper_estimate": 0.005358092422325499
     },
     {
       "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": 9686.23064118901,
-      "required_pages_per_second_upper_estimate": 0.004055422497968992
+      "required_token_positions_per_second": 9681.108550506158,
+      "required_pages_per_second_upper_estimate": 0.004053277985561498
     },
     {
       "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": 19372.46128237802,
-      "required_pages_per_second_upper_estimate": 0.0040347727989597185
+      "required_token_positions_per_second": 19362.217101012317,
+      "required_pages_per_second_upper_estimate": 0.004032639206138473
     },
     {
       "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": 29058.69192356703,
-      "required_pages_per_second_upper_estimate": 0.0017420242107643855
+      "required_token_positions_per_second": 29043.325651518477,
+      "required_pages_per_second_upper_estimate": 0.0017411030262180138
     },
     {
       "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": 29058.69192356703,
-      "required_pages_per_second_upper_estimate": 0.002259185481379542
+      "required_token_positions_per_second": 29043.325651518477,
+      "required_pages_per_second_upper_estimate": 0.002257990821316855
     },
     {
       "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": 29058.69192356703,
-      "required_pages_per_second_upper_estimate": 0.37273369920302496
+      "required_token_positions_per_second": 29043.325651518477,
+      "required_pages_per_second_upper_estimate": 0.3725365971642036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1042840292029723,
-    "producer_p95_seconds": 0.32765577599639073,
-    "gpu_update_seconds": 0.9514537017950578,
-    "producer_batches_per_second": 9.589196041262069,
-    "gpu_batches_per_second": 1.0510232900595713,
-    "producer_headroom": 9.123676070697309,
+    "producer_batch_seconds": 0.08808765961708787,
+    "producer_p95_seconds": 0.3102987239981303,
+    "gpu_update_seconds": 0.9519570978798868,
+    "producer_batches_per_second": 11.352327946354167,
+    "gpu_batches_per_second": 1.0504675076504078,
+    "producer_headroom": 10.806929165992049,
     "sustainable": true,
-    "buffer_seconds": 60.8930369148837,
+    "buffer_seconds": 60.925254264312755,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384556.8026664586,
-    "upload_bytes_per_second": 5005834.0311376825,
+    "generation_bytes_per_second": 384902.93525679735,
+    "upload_bytes_per_second": 4917912.886723278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.51693782801932,
+    "estimated_full_shards_per_hour": 82.5912098243517,
     "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": 1.4572524590003013,
-      "bytes": 10153840,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8581832199997734,
       "bytes": 2650389,
@@ -562,6 +556,12 @@
       "bytes": 17287207,
       "pages": 44,
       "shards": 5
+    },
+    {
+      "seconds": 1.2279312789978576,
+      "bytes": 1252954,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 684500, "loss": 0.5723017454147339, "lr": 0.0001, "grad_norm": 0.19126483496215455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100193504742, "tokens_per_second": 152234.8185075767, "data_wait_s": 1.79049966391176e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684600, "loss": 0.497664213180542, "lr": 0.0001, "grad_norm": 0.21434360141424869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100208048095, "tokens_per_second": 151214.19391769977, "data_wait_s": 1.1353004083503038e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684700, "loss": 0.59892338514328, "lr": 0.0001, "grad_norm": 0.2340129714638774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100222432027, "tokens_per_second": 149699.38474605192, "data_wait_s": 1.073999737855047e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 684800, "loss": 0.565108060836792, "lr": 0.0001, "grad_norm": 0.22957951672422106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100237080110, "tokens_per_second": 152459.09059993987, "data_wait_s": 9.927003702614456e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6157d8a27a0b495f752a3661f8238624245fc6e0ac5d6f3d6e883e2dd9ec5fc
-size 846492
+oid sha256:d667d8c61f98ea31fecbe06689efd16dec71e1244ab13c11723735e51108761f
+size 848610