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": 76055.0254187549,
-      "required_pages_per_second_upper_estimate": 0.005261663130441529
+      "required_token_positions_per_second": 76281.65135291884,
+      "required_pages_per_second_upper_estimate": 0.0052773416384116045
     },
     {
       "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": 9506.878177344362,
-      "required_pages_per_second_upper_estimate": 0.0039803313666625754
+      "required_token_positions_per_second": 9535.206419114855,
+      "required_pages_per_second_upper_estimate": 0.003992191809930924
     },
     {
       "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": 19013.756354688725,
-      "required_pages_per_second_upper_estimate": 0.003960064022207118
+      "required_token_positions_per_second": 19070.41283822971,
+      "required_pages_per_second_upper_estimate": 0.003971864073596758
     },
     {
       "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": 28520.634532033087,
-      "required_pages_per_second_upper_estimate": 0.0017097684917079924
+      "required_token_positions_per_second": 28605.619257344566,
+      "required_pages_per_second_upper_estimate": 0.0017148631962262536
     },
     {
       "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": 28520.634532033087,
-      "required_pages_per_second_upper_estimate": 0.0022173538858521265
+      "required_token_positions_per_second": 28605.619257344566,
+      "required_pages_per_second_upper_estimate": 0.0022239610744367927
     },
     {
       "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": 28520.634532033087,
-      "required_pages_per_second_upper_estimate": 0.3658320767054436
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28605.619257344566,
+      "required_pages_per_second_upper_estimate": 0.36692216951225054
     }
   ],
   "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": 0.07960520502339818,
-    "producer_p95_seconds": 0.2790929030015832,
-    "gpu_update_seconds": 0.9694033970018099,
-    "producer_batches_per_second": 12.561992644903963,
-    "gpu_batches_per_second": 1.031562302229206,
-    "producer_headroom": 12.177638343081652,
+    "producer_batch_seconds": 0.15194794976559933,
+    "producer_p95_seconds": 0.27668988200093736,
+    "gpu_update_seconds": 0.9665233865860571,
+    "producer_batches_per_second": 6.5812010069411135,
+    "gpu_batches_per_second": 1.0346361131852055,
+    "producer_headroom": 6.3608846850322935,
     "sustainable": true,
-    "buffer_seconds": 60.103010614112215,
+    "buffer_seconds": 60.8909733549216,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373417.7586702491,
-    "upload_bytes_per_second": 5246324.304168481,
+    "generation_bytes_per_second": 449831.6136649108,
+    "upload_bytes_per_second": 5246914.687523604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.12675829010585,
+    "estimated_full_shards_per_hour": 96.52339274845593,
     "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": 1.0907142409996595,
-      "bytes": 6115368,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3705160650024482,
       "bytes": 13033658,
@@ -562,6 +556,12 @@
       "bytes": 934829,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.0800791440015018,
+      "bytes": 6107947,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -307,3 +307,4 @@
 {"step": 750700, "loss": 0.6294246315956116, "lr": 5e-05, "grad_norm": 0.27773659105765264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109881302376, "tokens_per_second": 151934.11129469165, "data_wait_s": 1.4614000974688679e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750800, "loss": 0.5326976776123047, "lr": 5e-05, "grad_norm": 0.23603058862963358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109896032521, "tokens_per_second": 151875.49988644582, "data_wait_s": 1.4716002624481916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750900, "loss": 0.5826548933982849, "lr": 5e-05, "grad_norm": 0.24639832661260694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109910790107, "tokens_per_second": 152329.28974206263, "data_wait_s": 1.2206997780594975e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 751000, "loss": 0.5846525430679321, "lr": 5e-05, "grad_norm": 0.26256670313217834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109925231035, "tokens_per_second": 148437.5651073467, "data_wait_s": 1.0769999789772555e-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 sha256:7cb4ddafd6795084f638319f3b98f7dc011919e2f4331703b04ee3558cfcecc0
-size 659325
+oid sha256:61e7e322ad487c3441f99dff98412ca65734347cc13f38733360f6faf8710da7
+size 661443