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": 75248.6696748232,
-      "required_pages_per_second_upper_estimate": 0.006522363971403995
+      "required_token_positions_per_second": 74676.27376634613,
+      "required_pages_per_second_upper_estimate": 0.006472750144781369
     },
     {
       "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": 9406.0837093529,
-      "required_pages_per_second_upper_estimate": 0.0037522274251447665
+      "required_token_positions_per_second": 9334.534220793266,
+      "required_pages_per_second_upper_estimate": 0.0037236852643981436
     },
     {
       "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": 18812.1674187058,
-      "required_pages_per_second_upper_estimate": 0.004351863849478901
+      "required_token_positions_per_second": 18669.068441586533,
+      "required_pages_per_second_upper_estimate": 0.00431876041957834
     },
     {
       "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": 28218.251128058702,
-      "required_pages_per_second_upper_estimate": 0.0017183434439769915
+      "required_token_positions_per_second": 28003.6026623798,
+      "required_pages_per_second_upper_estimate": 0.0017052724785906104
     },
     {
       "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": 28218.251128058702,
-      "required_pages_per_second_upper_estimate": 0.0021723190849565433
+      "required_token_positions_per_second": 28003.6026623798,
+      "required_pages_per_second_upper_estimate": 0.0021557948518835988
     },
     {
       "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": 28218.251128058702,
-      "required_pages_per_second_upper_estimate": 0.3503762385991371
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28003.6026623798,
+      "required_pages_per_second_upper_estimate": 0.34771102303760754
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24027451208587536,
-    "producer_p95_seconds": 0.3866118180012563,
-    "gpu_update_seconds": 0.9797914078561845,
-    "producer_batches_per_second": 4.161906276778099,
-    "gpu_batches_per_second": 1.020625402490549,
-    "producer_headroom": 4.077800010289905,
+    "producer_batch_seconds": 0.11668934782028373,
+    "producer_p95_seconds": 0.37897779799823184,
+    "gpu_update_seconds": 0.9873015387817398,
+    "producer_batches_per_second": 8.569762524854674,
+    "gpu_batches_per_second": 1.0128617861103804,
+    "producer_headroom": 8.460939727783108,
     "sustainable": true,
-    "buffer_seconds": 60.74706728708344,
+    "buffer_seconds": 60.22539386568612,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430981.71113631636,
+    "generation_bytes_per_second": 430789.46280473354,
     "upload_bytes_per_second": 5879434.878421239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.47864246909253,
+    "estimated_full_shards_per_hour": 92.43739045244698,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 115900, "loss": 0.5962846651673317, "lr": 0.0003, "grad_norm": 0.14476901545958598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16960422649, "tokens_per_second": 148634.60798728914, "data_wait_s": 1.1573996744118631e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116000, "loss": 0.6011932659894228, "lr": 0.0003, "grad_norm": 0.1298500720331202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16975010873, "tokens_per_second": 148774.85320683385, "data_wait_s": 1.6281002899631858e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 116100, "loss": 0.5674298200756311, "lr": 0.0003, "grad_norm": 0.14025242568622506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16989562756, "tokens_per_second": 148254.01830830792, "data_wait_s": 8.544993761461228e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 116200, "loss": 0.63331014290452, "lr": 0.0003, "grad_norm": 0.14362439571118016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17004101407, "tokens_per_second": 148138.27393720526, "data_wait_s": 1.3131000741850585e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64468c060272e1edb44d2f99a7db3f9d6413d4fe2d5f7c33343bfcc4417f08ea
-size 663561
+oid sha256:37800f58e1ec0431516cd2ad255b7193c48df824f460ced4f51273ce29e66806
+size 665679