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": 76791.18722194068,
-      "required_pages_per_second_upper_estimate": 0.005312592512116639
+      "required_token_positions_per_second": 76744.25026710107,
+      "required_pages_per_second_upper_estimate": 0.005309345304672095
     },
     {
       "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": 9598.898402742585,
-      "required_pages_per_second_upper_estimate": 0.004018858313435993
+      "required_token_positions_per_second": 9593.031283387634,
+      "required_pages_per_second_upper_estimate": 0.004016401873081459
     },
     {
       "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": 19197.79680548517,
-      "required_pages_per_second_upper_estimate": 0.003998394794634949
+      "required_token_positions_per_second": 19186.06256677527,
+      "required_pages_per_second_upper_estimate": 0.003995950862164361
     },
     {
       "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": 28796.695208227753,
-      "required_pages_per_second_upper_estimate": 0.001726317907725617
+      "required_token_positions_per_second": 28779.093850162903,
+      "required_pages_per_second_upper_estimate": 0.0017252627331853466
     },
     {
       "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": 28796.695208227753,
-      "required_pages_per_second_upper_estimate": 0.0022388163891636746
+      "required_token_positions_per_second": 28779.093850162903,
+      "required_pages_per_second_upper_estimate": 0.0022374479609943245
     },
     {
       "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": 28796.695208227753,
-      "required_pages_per_second_upper_estimate": 0.3693730866488084
+      "required_token_positions_per_second": 28779.093850162903,
+      "required_pages_per_second_upper_estimate": 0.36914731532641837
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1757843944140376,
-    "producer_p95_seconds": 0.24998946199957572,
-    "gpu_update_seconds": 0.9601101723679893,
-    "producer_batches_per_second": 5.688787126601399,
-    "gpu_batches_per_second": 1.041547135714256,
-    "producer_headroom": 5.461862388686068,
+    "producer_batch_seconds": 0.08183375995308495,
+    "producer_p95_seconds": 0.21082608899996558,
+    "gpu_update_seconds": 0.9606973778934148,
+    "producer_batches_per_second": 12.219895561114349,
+    "gpu_batches_per_second": 1.0409105125203597,
+    "producer_headroom": 11.739621623693935,
     "sustainable": true,
-    "buffer_seconds": 60.48694085918333,
+    "buffer_seconds": 60.52393480728513,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472055.093985779,
-    "upload_bytes_per_second": 5538470.300464128,
+    "generation_bytes_per_second": 488764.05575074186,
+    "upload_bytes_per_second": 5607432.407656947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.29203428916956,
+    "estimated_full_shards_per_hour": 104.87738851920787,
     "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": 1.3278628770003706,
-      "bytes": 6286061,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9090305769996121,
       "bytes": 7087614,
@@ -562,6 +556,12 @@
       "bytes": 8291174,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.122991837999507,
+      "bytes": 10369739,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 412100, "loss": 0.6097952127456665, "lr": 0.0003, "grad_norm": 0.14925702554629555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60320879151, "tokens_per_second": 150830.78593537412, "data_wait_s": 1.4716000805492513e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412200, "loss": 0.595787525177002, "lr": 0.0003, "grad_norm": 0.16429479578764294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60335480850, "tokens_per_second": 151121.78452627858, "data_wait_s": 1.2638000043807551e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 412300, "loss": 0.6107044816017151, "lr": 0.0003, "grad_norm": 0.1553087075550869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60350193167, "tokens_per_second": 152804.51002770357, "data_wait_s": 1.5322000763262622e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 412400, "loss": 0.6050095558166504, "lr": 0.0003, "grad_norm": 0.1712635682673306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60364855851, "tokens_per_second": 151022.2995741828, "data_wait_s": 1.5658000847906806e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cd55041b437b78ca7b2cda4a543397188496b72b7f998c0013ad9c4709d59e5
-size 324450
+oid sha256:c853c12c8743af74872028d3c0db677d60d67015a72484aabade26b09e7f4a38
+size 326568