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": 75628.04578451069,
-      "required_pages_per_second_upper_estimate": 0.0065552473310717466
+      "required_token_positions_per_second": 76109.32094198161,
+      "required_pages_per_second_upper_estimate": 0.006596963041940592
     },
     {
       "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": 9453.505723063836,
-      "required_pages_per_second_upper_estimate": 0.0037711447754363477
+      "required_token_positions_per_second": 9513.665117747702,
+      "required_pages_per_second_upper_estimate": 0.003795143257438847
     },
     {
       "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": 18907.01144612767,
-      "required_pages_per_second_upper_estimate": 0.004373804346025079
+      "required_token_positions_per_second": 19027.330235495403,
+      "required_pages_per_second_upper_estimate": 0.004401637980406933
     },
     {
       "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": 28360.517169191506,
-      "required_pages_per_second_upper_estimate": 0.001727006699469747
+      "required_token_positions_per_second": 28540.995353243103,
+      "required_pages_per_second_upper_estimate": 0.0017379968739826377
     },
     {
       "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": 28360.517169191506,
-      "required_pages_per_second_upper_estimate": 0.002183271118620554
+      "required_token_positions_per_second": 28540.995353243103,
+      "required_pages_per_second_upper_estimate": 0.002197164828824435
     },
     {
       "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": 28360.517169191506,
-      "required_pages_per_second_upper_estimate": 0.35214270669619563
+      "required_token_positions_per_second": 28540.995353243103,
+      "required_pages_per_second_upper_estimate": 0.35438364172049
     }
   ],
   "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": 0.12245626577328039,
-    "producer_p95_seconds": 0.4018576669986942,
-    "gpu_update_seconds": 0.9748764394901259,
-    "producer_batches_per_second": 8.16618074775719,
-    "gpu_batches_per_second": 1.0257710202977253,
-    "producer_headroom": 7.961017211606343,
+    "producer_batch_seconds": 0.10242133557784427,
+    "producer_p95_seconds": 0.39289101699978346,
+    "gpu_update_seconds": 0.9687118356528645,
+    "producer_batches_per_second": 9.763590704594556,
+    "gpu_batches_per_second": 1.0322987323944988,
+    "producer_headroom": 9.458105874011036,
     "sustainable": true,
-    "buffer_seconds": 60.44233924838781,
+    "buffer_seconds": 60.060133810477595,
     "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": 238624.00864898262,
+    "generation_bytes_per_second": 252254.02117479703,
     "upload_bytes_per_second": 5891571.482619043,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.203157373448455,
+    "estimated_full_shards_per_hour": 54.12784077103551,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 152500, "loss": 0.6430874671787024, "lr": 0.0003, "grad_norm": 0.15707740170207865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22317954706, "tokens_per_second": 149912.69458676706, "data_wait_s": 1.2789001630153507e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152600, "loss": 0.628054790198803, "lr": 0.0003, "grad_norm": 0.1676492831230582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22332668751, "tokens_per_second": 151222.65423052188, "data_wait_s": 1.2320000678300858e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 152700, "loss": 0.621080506592989, "lr": 0.0003, "grad_norm": 0.1435728370251433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22347369001, "tokens_per_second": 150385.1652976462, "data_wait_s": 1.1670999811030924e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 152800, "loss": 0.6182908602058887, "lr": 0.0003, "grad_norm": 0.13115770544587926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22362109494, "tokens_per_second": 150142.12544732486, "data_wait_s": 1.6645004507154226e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c7edc98f6e71f7ac4162af474a05094ca8d0dafae1bb12d343d12a398f48bb9
-size 804132
+oid sha256:8ab04a3b06dc4ec2e81d4041cd9c017a5cc1ce2d8442a4134d1d1bea480d7ddd
+size 806250