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": 75415.54602183588,
-      "required_pages_per_second_upper_estimate": 0.005217422461966042
+      "required_token_positions_per_second": 76240.65072560521,
+      "required_pages_per_second_upper_estimate": 0.005274505119879485
     },
     {
       "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": 9426.943252729485,
-      "required_pages_per_second_upper_estimate": 0.003946864282957367
+      "required_token_positions_per_second": 9530.081340700652,
+      "required_pages_per_second_upper_estimate": 0.0039900460466227045
     },
     {
       "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": 18853.88650545897,
-      "required_pages_per_second_upper_estimate": 0.003926767348664002
+      "required_token_positions_per_second": 19060.162681401303,
+      "required_pages_per_second_upper_estimate": 0.003969729236244215
     },
     {
       "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": 28280.82975818845,
-      "required_pages_per_second_upper_estimate": 0.0016953925616766947
+      "required_token_positions_per_second": 28590.244022101957,
+      "required_pages_per_second_upper_estimate": 0.001713941474350084
     },
     {
       "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": 28280.82975818845,
-      "required_pages_per_second_upper_estimate": 0.0021987101194754326
+      "required_token_positions_per_second": 28590.244022101957,
+      "required_pages_per_second_upper_estimate": 0.0022227657175251923
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28280.82975818845,
-      "required_pages_per_second_upper_estimate": 0.36275611854886997
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28590.244022101957,
+      "required_pages_per_second_upper_estimate": 0.36672495250319975
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1504772235078633,
-    "producer_p95_seconds": 0.2783166330009408,
-    "gpu_update_seconds": 0.9776233666551024,
-    "producer_batches_per_second": 6.6455239981733465,
-    "gpu_batches_per_second": 1.0228888078048486,
-    "producer_headroom": 6.496819544281504,
+    "producer_batch_seconds": 0.09314583064860926,
+    "producer_p95_seconds": 0.26800784599981853,
+    "gpu_update_seconds": 0.9670431626475959,
+    "producer_batches_per_second": 10.735853586109286,
+    "gpu_batches_per_second": 1.0340800065864424,
+    "producer_headroom": 10.382033805632659,
     "sustainable": true,
-    "buffer_seconds": 60.61264873261635,
+    "buffer_seconds": 60.92371924679854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323275.0247687541,
-    "upload_bytes_per_second": 5259928.28034438,
+    "generation_bytes_per_second": 322655.00911066623,
+    "upload_bytes_per_second": 5223693.01142689,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.3672948579499,
+    "estimated_full_shards_per_hour": 69.23425393095008,
     "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": 1.2136837799989735,
-      "bytes": 9835275,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2959440570011793,
       "bytes": 5942943,
@@ -562,6 +556,12 @@
       "bytes": 7100100,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0310441840010753,
+      "bytes": 5933626,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 750300, "loss": 0.5430476665496826, "lr": 5e-05, "grad_norm": 0.24048518397270588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109822774191, "tokens_per_second": 149618.8203046912, "data_wait_s": 1.6566002159379423e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750400, "loss": 0.573595404624939, "lr": 5e-05, "grad_norm": 0.2674222553572067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109837321654, "tokens_per_second": 149834.9871041493, "data_wait_s": 1.664199953665957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750500, "loss": 0.5602898001670837, "lr": 5e-05, "grad_norm": 0.2433584197934176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109851982207, "tokens_per_second": 151543.01189973668, "data_wait_s": 1.3510998542187735e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 750600, "loss": 0.4930821657180786, "lr": 5e-05, "grad_norm": 0.21799323546020904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109866588196, "tokens_per_second": 150365.27921716875, "data_wait_s": 1.4193999959388748e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d74281ba85f0d6f10ee34b6028ea0b4dfa64fe2d6acd60ce00e5d5fab71b2024
-size 650853
+oid sha256:0856f6f82d3252a2dc18d26a8e860af5294bf3a1395b1b42291fd357a715b08a
+size 652971