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": 75237.04493940083,
-      "required_pages_per_second_upper_estimate": 0.006521356368800206
+      "required_token_positions_per_second": 75391.27106636859,
+      "required_pages_per_second_upper_estimate": 0.006534724298603226
     },
     {
       "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": 9404.630617425104,
-      "required_pages_per_second_upper_estimate": 0.003751647765049108
+      "required_token_positions_per_second": 9423.908883296073,
+      "required_pages_per_second_upper_estimate": 0.0037593381535407982
     },
     {
       "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": 18809.261234850208,
-      "required_pages_per_second_upper_estimate": 0.004351191554990998
+      "required_token_positions_per_second": 18847.817766592147,
+      "required_pages_per_second_upper_estimate": 0.004360110930037713
     },
     {
       "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": 28213.891852275312,
-      "required_pages_per_second_upper_estimate": 0.0017180779869531355
+      "required_token_positions_per_second": 28271.726649888216,
+      "required_pages_per_second_upper_estimate": 0.001721599822692029
     },
     {
       "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": 28213.891852275312,
-      "required_pages_per_second_upper_estimate": 0.0021719834958394903
+      "required_token_positions_per_second": 28271.726649888216,
+      "required_pages_per_second_upper_estimate": 0.002176435778656698
     },
     {
       "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": 28213.891852275312,
-      "required_pages_per_second_upper_estimate": 0.3503221109834649
+      "required_token_positions_per_second": 28271.726649888216,
+      "required_pages_per_second_upper_estimate": 0.35104022560920095
     }
   ],
   "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": 0.11157404205468424,
-    "producer_p95_seconds": 0.3871487979995436,
-    "gpu_update_seconds": 0.979942793598336,
-    "producer_batches_per_second": 8.962658173752313,
-    "gpu_batches_per_second": 1.020467731925467,
-    "producer_headroom": 8.782892288853802,
+    "producer_batch_seconds": 0.20149608757839133,
+    "producer_p95_seconds": 0.3770669419973274,
+    "gpu_update_seconds": 0.9779381479733328,
+    "producer_batches_per_second": 4.962875517922668,
+    "gpu_batches_per_second": 1.0225595576493134,
+    "producer_headroom": 4.853385292619488,
     "sustainable": true,
-    "buffer_seconds": 60.75645320309683,
+    "buffer_seconds": 60.63216517434663,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418044.68885636335,
+    "generation_bytes_per_second": 439262.3291501913,
     "upload_bytes_per_second": 5361384.650076382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.70265864628006,
+    "estimated_full_shards_per_hour": 94.25547033194832,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 227200, "loss": 0.5655167624354362, "lr": 0.0003, "grad_norm": 0.15623009229924723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33255759918, "tokens_per_second": 146967.80869872757, "data_wait_s": 1.4302997442428023e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227300, "loss": 0.5979587100446224, "lr": 0.0003, "grad_norm": 0.14057180409488337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33270325224, "tokens_per_second": 148665.90845497118, "data_wait_s": 1.1363998055458069e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227400, "loss": 0.6005997657775879, "lr": 0.0003, "grad_norm": 0.1638793844328349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33285046722, "tokens_per_second": 149696.28761427378, "data_wait_s": 1.3261997082736343e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 227500, "loss": 0.674255259335041, "lr": 0.0003, "grad_norm": 0.20765405607166754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33299770332, "tokens_per_second": 149994.43463273006, "data_wait_s": 1.1252996046096087e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25eef2e2fa1abbba4b6f785d157e0576deac5b0732cfd173495ed4de5a7cdab8
-size 691095
+oid sha256:44e3453eda62d38743402177e0c443e6922afc522588c7d17dc4d1558133f56f
+size 693213