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": 77540.29240535131,
-      "required_pages_per_second_upper_estimate": 0.005364417346868484
+      "required_token_positions_per_second": 77403.41268789848,
+      "required_pages_per_second_upper_estimate": 0.005354947690410395
     },
     {
       "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": 9692.536550668914,
-      "required_pages_per_second_upper_estimate": 0.004058062650585863
+      "required_token_positions_per_second": 9675.42658598731,
+      "required_pages_per_second_upper_estimate": 0.004050899065670369
     },
     {
       "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": 19385.073101337828,
-      "required_pages_per_second_upper_estimate": 0.004037399508253015
+      "required_token_positions_per_second": 19350.85317197462,
+      "required_pages_per_second_upper_estimate": 0.004030272399406882
     },
     {
       "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": 29077.60965200674,
-      "required_pages_per_second_upper_estimate": 0.0017431582996986475
+      "required_token_positions_per_second": 29026.27975796193,
+      "required_pages_per_second_upper_estimate": 0.0017400811509268711
     },
     {
       "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": 29077.60965200674,
-      "required_pages_per_second_upper_estimate": 0.002260656251555436
+      "required_token_positions_per_second": 29026.27975796193,
+      "required_pages_per_second_upper_estimate": 0.0022566655780697953
     },
     {
       "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": 29077.60965200674,
-      "required_pages_per_second_upper_estimate": 0.37297635551117536
+      "required_token_positions_per_second": 29026.27975796193,
+      "required_pages_per_second_upper_estimate": 0.3723179507441147
     }
   ],
   "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": 0.057988607703109096,
-    "producer_p95_seconds": 0.18839511599981051,
-    "gpu_update_seconds": 0.9508346913960282,
-    "producer_batches_per_second": 17.24476650861863,
-    "gpu_batches_per_second": 1.051707525029179,
-    "producer_headroom": 16.396922241418956,
+    "producer_batch_seconds": 0.07595258480466782,
+    "producer_p95_seconds": 0.17903192800076795,
+    "gpu_update_seconds": 0.952516141598068,
+    "producer_batches_per_second": 13.166108863467448,
+    "gpu_batches_per_second": 1.0498509750420257,
+    "producer_headroom": 12.540931214490136,
     "sustainable": true,
-    "buffer_seconds": 60.8534202493458,
+    "buffer_seconds": 60.00851692067828,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287469.38891614845,
-    "upload_bytes_per_second": 4263491.428027692,
+    "generation_bytes_per_second": 298472.7689715035,
+    "upload_bytes_per_second": 4226793.635046923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.684238916524315,
+    "estimated_full_shards_per_hour": 64.04530813082532,
     "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": [
@@ -472,6 +472,12 @@
       "bytes": 11139901,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.9246388850006042,
+      "bytes": 1100937,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 935000, "loss": 0.5534255504608154, "lr": 3e-05, "grad_norm": 0.35179548294994867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136857750792, "tokens_per_second": 132722.11465712264, "data_wait_s": 1.3328000932233408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935100, "loss": 0.528160035610199, "lr": 3e-05, "grad_norm": 0.3339950495529836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136872314892, "tokens_per_second": 140259.65118426422, "data_wait_s": 1.6283998775179498e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935200, "loss": 0.5229005813598633, "lr": 3e-05, "grad_norm": 0.3512448534306371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136886870395, "tokens_per_second": 152078.88748921623, "data_wait_s": 1.5336001524701715e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 935300, "loss": 0.576909065246582, "lr": 3e-05, "grad_norm": 0.3603467603905492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136901412982, "tokens_per_second": 151158.33212154746, "data_wait_s": 1.6465999578940682e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9cf52ebabd28b3df2b3212e442957dae701dc7271a765da0d8d048221cbe3f8e
-size 217767
+oid sha256:0972d0f9a7ad47f826698d11b155042e13288889057c40d2264bf8aec34b8575
+size 219885