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": 77546.39792240928,
-      "required_pages_per_second_upper_estimate": 0.005364839740705305
+      "required_token_positions_per_second": 77328.7214001537,
+      "required_pages_per_second_upper_estimate": 0.005349780373816547
     },
     {
       "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": 9693.29974030116,
-      "required_pages_per_second_upper_estimate": 0.004058382182147673
+      "required_token_positions_per_second": 9666.090175019213,
+      "required_pages_per_second_upper_estimate": 0.004046990105364457
     },
     {
       "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": 19386.59948060232,
-      "required_pages_per_second_upper_estimate": 0.00403771741280048
+      "required_token_positions_per_second": 19332.180350038427,
+      "required_pages_per_second_upper_estimate": 0.004026383343033003
     },
     {
       "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": 29079.89922090348,
-      "required_pages_per_second_upper_estimate": 0.0017432955558580377
+      "required_token_positions_per_second": 28998.27052505764,
+      "required_pages_per_second_upper_estimate": 0.0017384020401818813
     },
     {
       "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": 29079.89922090348,
-      "required_pages_per_second_upper_estimate": 0.002260834255466408
+      "required_token_positions_per_second": 28998.27052505764,
+      "required_pages_per_second_upper_estimate": 0.002254487983411082
     },
     {
       "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": 29079.89922090348,
-      "required_pages_per_second_upper_estimate": 0.3730057236426352
+      "required_token_positions_per_second": 28998.27052505764,
+      "required_pages_per_second_upper_estimate": 0.3719586783783897
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.13331641442948694,
-    "producer_p95_seconds": 0.34906877399771474,
-    "gpu_update_seconds": 0.9507598286353692,
-    "producer_batches_per_second": 7.500951809118112,
-    "gpu_batches_per_second": 1.0517903364042058,
-    "producer_headroom": 7.131603656639299,
+    "producer_batch_seconds": 0.11340473017207842,
+    "producer_p95_seconds": 0.3232885629986413,
+    "gpu_update_seconds": 0.9534361704815858,
+    "producer_batches_per_second": 8.817974333897862,
+    "gpu_batches_per_second": 1.0488379096158,
+    "producer_headroom": 8.407375680316491,
     "sustainable": true,
-    "buffer_seconds": 60.84862903266363,
+    "buffer_seconds": 60.0664787403399,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 550408.0888667927,
+    "generation_bytes_per_second": 555690.4236854177,
     "upload_bytes_per_second": 3698903.8537535975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.10476302626454,
+    "estimated_full_shards_per_hour": 119.2382291118803,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 526300, "loss": 0.638948380947113, "lr": 0.0003, "grad_norm": 0.16214304313669964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77032790042, "tokens_per_second": 152668.93492124518, "data_wait_s": 1.5463003364857286e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526400, "loss": 0.539824366569519, "lr": 0.0003, "grad_norm": 0.15327712469377483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77047436084, "tokens_per_second": 153171.25112464483, "data_wait_s": 1.7212005332112312e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526500, "loss": 0.5812841653823853, "lr": 0.0003, "grad_norm": 0.17130374021271788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77062021391, "tokens_per_second": 151421.90567680768, "data_wait_s": 1.4544995792675763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526600, "loss": 0.5899205803871155, "lr": 0.0003, "grad_norm": 0.1611243733054155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77076707540, "tokens_per_second": 152968.71686749623, "data_wait_s": 1.605800207471475e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e076739a5e677b9944ed7416d5e5aa94067c782f9666fe0ac85bde3fb0c7d9d9
-size 885399
+oid sha256:24ae5ca326320d8dc154a91f0a77b88d6c4972fea3e27b3ca112f71e8dc64c62
+size 887517