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": 74727.71575576495,
-      "required_pages_per_second_upper_estimate": 0.006477209006045662
+      "required_token_positions_per_second": 74750.32258781989,
+      "required_pages_per_second_upper_estimate": 0.006479168508951692
     },
     {
       "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": 9340.964469470618,
-      "required_pages_per_second_upper_estimate": 0.003726250386736325
+      "required_token_positions_per_second": 9343.790323477486,
+      "required_pages_per_second_upper_estimate": 0.0037273776621499465
     },
     {
       "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": 18681.928938941237,
-      "required_pages_per_second_upper_estimate": 0.004321735469304327
+      "required_token_positions_per_second": 18687.58064695497,
+      "required_pages_per_second_upper_estimate": 0.004323042892486641
     },
     {
       "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": 28022.893408411852,
-      "required_pages_per_second_upper_estimate": 0.0017064471838132398
+      "required_token_positions_per_second": 28031.370970432454,
+      "required_pages_per_second_upper_estimate": 0.0017069634228619641
     },
     {
       "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": 28022.893408411852,
-      "required_pages_per_second_upper_estimate": 0.002157279906913349
+      "required_token_positions_per_second": 28031.370970432454,
+      "required_pages_per_second_upper_estimate": 0.0021579325331050887
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28022.893408411852,
-      "required_pages_per_second_upper_estimate": 0.34795054954135185
+      "required_token_positions_per_second": 28031.370970432454,
+      "required_pages_per_second_upper_estimate": 0.34805581248907275
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12381515139037447,
-    "producer_p95_seconds": 0.40632982500392245,
-    "gpu_update_seconds": 0.9866218879346942,
-    "producer_batches_per_second": 8.076555968882344,
-    "gpu_batches_per_second": 1.013559512746378,
-    "producer_headroom": 7.968506898028922,
+    "producer_batch_seconds": 0.3826237511561317,
+    "producer_p95_seconds": 0.4105375700019067,
+    "gpu_update_seconds": 0.9863235026628974,
+    "producer_batches_per_second": 2.613533522104708,
+    "gpu_batches_per_second": 1.013866137530109,
+    "producer_headroom": 2.577789537849214,
     "sustainable": true,
-    "buffer_seconds": 60.18393516401635,
+    "buffer_seconds": 60.165733662436736,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526500.4699304724,
-    "upload_bytes_per_second": 3783215.4001733093,
+    "generation_bytes_per_second": 461976.2499679752,
+    "upload_bytes_per_second": 3821985.769711382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.97474454341535,
+    "estimated_full_shards_per_hour": 99.12934898642962,
     "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": [
-    {
-      "seconds": 0.9801313110001502,
-      "bytes": 949416,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.2882471760021872,
       "bytes": 6045719,
@@ -562,6 +556,12 @@
       "bytes": 11930213,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.2040022809960647,
+      "bytes": 6093346,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 351100, "loss": 0.5874341055750847, "lr": 0.0003, "grad_norm": 0.1632072478115318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51391790571, "tokens_per_second": 148151.8317907659, "data_wait_s": 1.4305005606729537e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351200, "loss": 0.5624159909784794, "lr": 0.0003, "grad_norm": 0.15156771187556475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51406481924, "tokens_per_second": 149364.46151921395, "data_wait_s": 1.1673997505567968e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351300, "loss": 0.5609304569661617, "lr": 0.0003, "grad_norm": 0.15011317001339225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51421204613, "tokens_per_second": 149597.94732004078, "data_wait_s": 1.2988006346859038e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 351400, "loss": 0.6106679812073708, "lr": 0.0003, "grad_norm": 0.15877163517799867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51435954437, "tokens_per_second": 150004.92010822124, "data_wait_s": 1.2057003914378583e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0daf594aa489537e558de489402f05251025507088fe7b9cc1c8fd93237dd6d6
-size 774480
+oid sha256:4b164c249a84ccb3ef3108083270936abd8dab5fc3c6d46617b0edf7d95f03a8
+size 776598