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": 77794.62818775291,
-      "required_pages_per_second_upper_estimate": 0.005382012886435347
+      "required_token_positions_per_second": 77391.4821562272,
+      "required_pages_per_second_upper_estimate": 0.005354122308546728
     },
     {
       "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": 9724.328523469114,
-      "required_pages_per_second_upper_estimate": 0.004071373285705422
+      "required_token_positions_per_second": 9673.9352695284,
+      "required_pages_per_second_upper_estimate": 0.004050274682611251
     },
     {
       "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": 19448.657046938228,
-      "required_pages_per_second_upper_estimate": 0.004050642367300171
+      "required_token_positions_per_second": 19347.8705390568,
+      "required_pages_per_second_upper_estimate": 0.00402965119562742
     },
     {
       "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": 29172.98557040734,
-      "required_pages_per_second_upper_estimate": 0.0017488759403761684
+      "required_token_positions_per_second": 29021.8058085852,
+      "required_pages_per_second_upper_estimate": 0.0017398129444930613
     },
     {
       "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": 29172.98557040734,
-      "required_pages_per_second_upper_estimate": 0.0022680713096967535
+      "required_token_positions_per_second": 29021.8058085852,
+      "required_pages_per_second_upper_estimate": 0.002256317748184578
     },
     {
       "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": 29172.98557040734,
-      "required_pages_per_second_upper_estimate": 0.3741997353857357
+      "required_token_positions_per_second": 29021.8058085852,
+      "required_pages_per_second_upper_estimate": 0.3722605637252626
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09628913921122262,
-    "producer_p95_seconds": 0.23010016099942732,
-    "gpu_update_seconds": 0.94772610548458,
-    "producer_batches_per_second": 10.38538726373253,
-    "gpu_batches_per_second": 1.0551571748555897,
-    "producer_headroom": 9.842502625406391,
+    "producer_batch_seconds": 0.08700346036707174,
+    "producer_p95_seconds": 0.22561216200119816,
+    "gpu_update_seconds": 0.9526629797729953,
+    "producer_batches_per_second": 11.493795715491688,
+    "gpu_batches_per_second": 1.0496891568498699,
+    "producer_headroom": 10.949713675222398,
     "sustainable": true,
-    "buffer_seconds": 60.65447075101312,
+    "buffer_seconds": 60.017767725698704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309336.012710601,
+    "generation_bytes_per_second": 280870.7560220441,
     "upload_bytes_per_second": 5177278.0525495885,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.37630735386392,
+    "estimated_full_shards_per_hour": 60.26832590576165,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 901800, "loss": 0.5543702840805054, "lr": 3e-05, "grad_norm": 0.321183228617444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131997800547, "tokens_per_second": 151369.25784464725, "data_wait_s": 8.856000931700692e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 901900, "loss": 0.5142814517021179, "lr": 3e-05, "grad_norm": 0.36111615552078913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132012334157, "tokens_per_second": 152130.04555850985, "data_wait_s": 8.2970000221394e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 902000, "loss": 0.5099591016769409, "lr": 3e-05, "grad_norm": 0.31409916288570733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132026993992, "tokens_per_second": 134465.43752851879, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 902100, "loss": 0.5030158162117004, "lr": 3e-05, "grad_norm": 0.3099639529913761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132041691282, "tokens_per_second": 152593.38545743533, "data_wait_s": 1.7455000488553196e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:45837a9ef1ebcf8e5f6ff7d50f5519de01c66a99abcd33b06eaf11e11589cb44
-size 362574
+oid sha256:7f8deedc01ae110c2c9818fa166af827fed30fa444eb46d11ced5860e0a84a07
+size 364692