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": 75411.63119095686,
-      "required_pages_per_second_upper_estimate": 0.006536489062600273
+      "required_token_positions_per_second": 73545.61526196354,
+      "required_pages_per_second_upper_estimate": 0.006374747531249267
     },
     {
       "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": 9426.453898869608,
-      "required_pages_per_second_upper_estimate": 0.003760353398304455
+      "required_token_positions_per_second": 9193.201907745442,
+      "required_pages_per_second_upper_estimate": 0.003667305691616979
     },
     {
       "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": 18852.907797739215,
-      "required_pages_per_second_upper_estimate": 0.004361288419162099
+      "required_token_positions_per_second": 18386.403815490885,
+      "required_pages_per_second_upper_estimate": 0.004253370933058623
     },
     {
       "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": 28279.36169660882,
-      "required_pages_per_second_upper_estimate": 0.0017220647569793194
+      "required_token_positions_per_second": 27579.60572323633,
+      "required_pages_per_second_upper_estimate": 0.0016794532895367913
     },
     {
       "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": 28279.36169660882,
-      "required_pages_per_second_upper_estimate": 0.002177023545688412
+      "required_token_positions_per_second": 27579.60572323633,
+      "required_pages_per_second_upper_estimate": 0.002123154393809687
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28279.36169660882,
-      "required_pages_per_second_upper_estimate": 0.3511350273366132
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27579.60572323633,
+      "required_pages_per_second_upper_estimate": 0.34244640007991767
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.30460129902351696,
-    "producer_p95_seconds": 0.30531534100009594,
-    "gpu_update_seconds": 0.9776741178467075,
-    "producer_batches_per_second": 3.282980089729671,
-    "gpu_batches_per_second": 1.0228357095127611,
-    "producer_headroom": 3.2096846631347606,
+    "producer_batch_seconds": 0.2766273766950178,
+    "producer_p95_seconds": 0.38842329399994924,
+    "gpu_update_seconds": 1.002479885950873,
+    "producer_batches_per_second": 3.6149712004191903,
+    "gpu_batches_per_second": 0.9975262486702955,
+    "producer_headroom": 3.62393591671192,
     "sustainable": true,
-    "buffer_seconds": 60.61579530649587,
+    "buffer_seconds": 60.14879315705238,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455913.48611340876,
+    "generation_bytes_per_second": 438858.77645443176,
     "upload_bytes_per_second": 4972521.479172338,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.82842099715897,
+    "estimated_full_shards_per_hour": 94.16887731766428,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 384900, "loss": 0.5901133706793189, "lr": 0.0003, "grad_norm": 0.1846310036791288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56338883655, "tokens_per_second": 149288.21830866195, "data_wait_s": 1.2441996659617871e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385000, "loss": 0.570076484233141, "lr": 0.0003, "grad_norm": 0.14780607451161634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56353426047, "tokens_per_second": 148358.1146194847, "data_wait_s": 1.2352000339888036e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 385100, "loss": 0.6046611033380032, "lr": 0.0003, "grad_norm": 0.1473527156192229, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56367925588, "tokens_per_second": 129382.63438104208, "data_wait_s": 1.2508000509114936e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 385200, "loss": 0.540040971711278, "lr": 0.0003, "grad_norm": 0.14787405830661698, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56382360865, "tokens_per_second": 147281.05100936638, "data_wait_s": 1.2210999557282776e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03f7b032b161f3758cadc70cddeb04ae4324d159b327efe58d697d124f87c07c
-size 642381
+oid sha256:03b9dd492e72416512d97cfb8aeb57e3a7028b4795485596fea2d8dedeb8e830
+size 644499