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": 76228.63617593308,
-      "required_pages_per_second_upper_estimate": 0.006607304984017809
+      "required_token_positions_per_second": 76067.6166343916,
+      "required_pages_per_second_upper_estimate": 0.0065933482182572905
     },
     {
       "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": 9528.579521991634,
-      "required_pages_per_second_upper_estimate": 0.0038010928362819667
+      "required_token_positions_per_second": 9508.45207929895,
+      "required_pages_per_second_upper_estimate": 0.0037930636984597694
     },
     {
       "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": 19057.15904398327,
-      "required_pages_per_second_upper_estimate": 0.004408538350281501
+      "required_token_positions_per_second": 19016.9041585979,
+      "required_pages_per_second_upper_estimate": 0.0043992260910093815
     },
     {
       "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": 28585.7385659749,
-      "required_pages_per_second_upper_estimate": 0.0017407215008885093
+      "required_token_positions_per_second": 28525.35623789685,
+      "required_pages_per_second_upper_estimate": 0.0017370445339101477
     },
     {
       "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": 28585.7385659749,
-      "required_pages_per_second_upper_estimate": 0.002200609285197669
+      "required_token_positions_per_second": 28525.35623789685,
+      "required_pages_per_second_upper_estimate": 0.0021959608864332383
     },
     {
       "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": 28585.7385659749,
-      "required_pages_per_second_upper_estimate": 0.3549392026767188
+      "required_token_positions_per_second": 28525.35623789685,
+      "required_pages_per_second_upper_estimate": 0.3541894562486416
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.42674889275770056,
-    "producer_p95_seconds": 0.40972126799897524,
-    "gpu_update_seconds": 0.9671955802782344,
-    "producer_batches_per_second": 2.3432984056218276,
-    "gpu_batches_per_second": 1.0339170488272174,
-    "producer_headroom": 2.266427861190465,
+    "producer_batch_seconds": 0.09742869203904547,
+    "producer_p95_seconds": 0.36168286099564284,
+    "gpu_update_seconds": 0.9692429349320009,
+    "producer_batches_per_second": 10.26391691268154,
+    "gpu_batches_per_second": 1.0317330815211534,
+    "producer_headroom": 9.948228952345655,
     "sustainable": true,
-    "buffer_seconds": 60.933321557528764,
+    "buffer_seconds": 60.09306196578405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383030.5496539843,
-    "upload_bytes_per_second": 5746051.909421746,
+    "generation_bytes_per_second": 382789.2603462038,
+    "upload_bytes_per_second": 5686576.56180762,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.18943946089408,
+    "estimated_full_shards_per_hour": 82.13766439237199,
     "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": 1.4035626420009066,
-      "bytes": 10955791,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.4038716970026144,
       "bytes": 7663039,
@@ -562,6 +556,12 @@
       "bytes": 7464078,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1531412479962455,
+      "bytes": 5169545,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 314000, "loss": 0.5206610346212983, "lr": 0.0003, "grad_norm": 0.15408984745557255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45961037178, "tokens_per_second": 151764.0794378983, "data_wait_s": 1.0324001777917147e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314100, "loss": 0.5576486960053444, "lr": 0.0003, "grad_norm": 0.15534584045263586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45975665453, "tokens_per_second": 149673.9852173171, "data_wait_s": 1.1534997611306608e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314200, "loss": 0.5247049517929554, "lr": 0.0003, "grad_norm": 0.16029382198406003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45990174774, "tokens_per_second": 148771.5261618356, "data_wait_s": 1.0413998097646981e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 314300, "loss": 0.5914095975458622, "lr": 0.0003, "grad_norm": 0.1586909353515651, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46004678255, "tokens_per_second": 148714.66425481616, "data_wait_s": 1.3215998478699476e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d67c4cb0c32522b7c9c60904bf0b585f8b01a2984ae545463d57520c7083b57
-size 724983
+oid sha256:fc5e198968fb1b92dfbac0032e39d9cea06aa512fb4bb8ebd8c150c02edba873
+size 727101