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": 76358.15403710921,
-      "required_pages_per_second_upper_estimate": 0.005282634271614578
+      "required_token_positions_per_second": 77579.90374283993,
+      "required_pages_per_second_upper_estimate": 0.005367157751106913
     },
     {
       "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": 9544.769254638652,
-      "required_pages_per_second_upper_estimate": 0.003996195569470024
+      "required_token_positions_per_second": 9697.487967854991,
+      "required_pages_per_second_upper_estimate": 0.004060135705564326
     },
     {
       "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": 19089.538509277303,
-      "required_pages_per_second_upper_estimate": 0.003975847446498109
+      "required_token_positions_per_second": 19394.975935709983,
+      "required_pages_per_second_upper_estimate": 0.00403946200749743
     },
     {
       "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": 28634.307763915953,
-      "required_pages_per_second_upper_estimate": 0.0017165830283903951
+      "required_token_positions_per_second": 29092.463903564974,
+      "required_pages_per_second_upper_estimate": 0.0017440487893984346
     },
     {
       "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": 28634.307763915953,
-      "required_pages_per_second_upper_estimate": 0.0022261914796353134
+      "required_token_positions_per_second": 29092.463903564974,
+      "required_pages_per_second_upper_estimate": 0.002261811104277141
     },
     {
       "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": 28634.307763915953,
-      "required_pages_per_second_upper_estimate": 0.3672901548712299
+      "required_token_positions_per_second": 29092.463903564974,
+      "required_pages_per_second_upper_estimate": 0.3731668899009117
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12204765596874267,
-    "producer_p95_seconds": 0.32526555199729046,
-    "gpu_update_seconds": 0.9655550337711021,
-    "producer_batches_per_second": 8.193520736327025,
-    "gpu_batches_per_second": 1.0356737472481177,
-    "producer_headroom": 7.911295191268467,
+    "producer_batch_seconds": 0.11659131987505589,
+    "producer_p95_seconds": 0.3559511380008189,
+    "gpu_update_seconds": 0.9503492069852506,
+    "producer_batches_per_second": 8.576967831495875,
+    "gpu_batches_per_second": 1.05224478817871,
+    "producer_headroom": 8.15111457700011,
     "sustainable": true,
-    "buffer_seconds": 60.82996712757943,
+    "buffer_seconds": 60.82234924705604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353061.3498725051,
-    "upload_bytes_per_second": 5000542.364859272,
+    "generation_bytes_per_second": 362966.8801118111,
+    "upload_bytes_per_second": 4972988.5449049985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.75874683505407,
+    "estimated_full_shards_per_hour": 77.88424303546667,
     "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": 1.4597645020003256,
-      "bytes": 6256363,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2800013509986456,
       "bytes": 667279,
@@ -562,6 +556,12 @@
       "bytes": 9959393,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2251598969960469,
+      "bytes": 2667726,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -404,3 +404,4 @@
 {"step": 685400, "loss": 0.5695536136627197, "lr": 0.0001, "grad_norm": 0.1888986239217438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100324882735, "tokens_per_second": 153663.03866914738, "data_wait_s": 1.4646997442469e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685500, "loss": 0.554060161113739, "lr": 0.0001, "grad_norm": 0.20678600802984906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100339543873, "tokens_per_second": 153235.66768583274, "data_wait_s": 8.188995707314461e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 685600, "loss": 0.5104237198829651, "lr": 0.0001, "grad_norm": 0.19530410664371448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100354050683, "tokens_per_second": 151499.80442519503, "data_wait_s": 1.3083001249469817e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685700, "loss": 0.5354567170143127, "lr": 0.0001, "grad_norm": 0.20256810840400116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100368548106, "tokens_per_second": 151211.0827327131, "data_wait_s": 1.2427000910975039e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2df5c3aab5d52c61fc22856a1a58bc937a267dc71d3bbe35f149a70b8a302d22
-size 866337
+oid sha256:cddbbe62e2e1c2ddffe633d0957ae51266db8492d3f83aac5fd03d0adffc6f7b
+size 868455