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": 77470.29884396623,
-      "required_pages_per_second_upper_estimate": 0.005359575029884426
+      "required_token_positions_per_second": 77157.1717398911,
+      "required_pages_per_second_upper_estimate": 0.005337912170269527
     },
     {
       "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": 9683.787355495779,
-      "required_pages_per_second_upper_estimate": 0.0040543995452708426
+      "required_token_positions_per_second": 9644.646467486387,
+      "required_pages_per_second_upper_estimate": 0.004038012072815997
     },
     {
       "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": 19367.574710991557,
-      "required_pages_per_second_upper_estimate": 0.0040337550550074725
+      "required_token_positions_per_second": 19289.292934972775,
+      "required_pages_per_second_upper_estimate": 0.00401745102549202
     },
     {
       "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": 29051.362066487334,
-      "required_pages_per_second_upper_estimate": 0.0017415847970245015
+      "required_token_positions_per_second": 28933.93940245916,
+      "required_pages_per_second_upper_estimate": 0.001734545487610039
     },
     {
       "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": 29051.362066487334,
-      "required_pages_per_second_upper_estimate": 0.0022586156172322296
+      "required_token_positions_per_second": 28933.93940245916,
+      "required_pages_per_second_upper_estimate": 0.002249486521591754
     },
     {
       "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": 29051.362066487334,
-      "required_pages_per_second_upper_estimate": 0.372639679666594
+      "required_token_positions_per_second": 28933.93940245916,
+      "required_pages_per_second_upper_estimate": 0.37113350781107424
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0999953806952476,
-    "producer_p95_seconds": 0.3017256009989069,
-    "gpu_update_seconds": 0.9516937600627612,
-    "producer_batches_per_second": 10.000461951814202,
-    "gpu_batches_per_second": 1.050758176594594,
-    "producer_headroom": 9.517377237286638,
+    "producer_batch_seconds": 0.08941187418713525,
+    "producer_p95_seconds": 0.30214533999969717,
+    "gpu_update_seconds": 0.9555560207487728,
+    "producer_batches_per_second": 11.18419683169869,
+    "gpu_batches_per_second": 1.0465111184338527,
+    "producer_headroom": 10.687126619769032,
     "sustainable": true,
-    "buffer_seconds": 60.908400644016716,
+    "buffer_seconds": 60.200029307172684,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375113.03417912155,
-    "upload_bytes_per_second": 4900715.187103541,
+    "generation_bytes_per_second": 376986.3101951353,
+    "upload_bytes_per_second": 4897430.65850128,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.4905249503158,
+    "estimated_full_shards_per_hour": 80.89248637571853,
     "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": 0.9852680410003813,
-      "bytes": 917014,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 3.7047144000025582,
       "bytes": 6068293,
@@ -562,6 +556,12 @@
       "bytes": 6246238,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.040108234003128,
+      "bytes": 901723,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -357,3 +357,4 @@
 {"step": 920700, "loss": 0.5966166257858276, "lr": 3e-05, "grad_norm": 0.4249011900687419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134763610364, "tokens_per_second": 153825.9132291072, "data_wait_s": 1.3163000403437763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 920800, "loss": 0.5239759683609009, "lr": 3e-05, "grad_norm": 0.3302813690938352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134778297799, "tokens_per_second": 153614.77644577468, "data_wait_s": 1.9553997844923288e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 920900, "loss": 0.5707686543464661, "lr": 3e-05, "grad_norm": 0.35707305900982755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134793014607, "tokens_per_second": 153371.9813734807, "data_wait_s": 1.4352001016959548e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 921000, "loss": 0.5189375281333923, "lr": 3e-05, "grad_norm": 0.33550743073941924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134807539492, "tokens_per_second": 151413.46363908608, "data_wait_s": 1.3078999472782016e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fa6f3af525b18edb66d1b996d379d147eacd5730980d6dc2ce505ddc11bafe4
-size 766008
+oid sha256:02d8fa14cfd5cc59ac9892a1a80a060e6ee5572ebdbaf9973deccc648eefcdde
+size 768126