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": 75838.97141279065,
-      "required_pages_per_second_upper_estimate": 0.00524671601299456
+      "required_token_positions_per_second": 76624.6568465838,
+      "required_pages_per_second_upper_estimate": 0.005301071554345743
     },
     {
       "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": 9479.871426598831,
-      "required_pages_per_second_upper_estimate": 0.003969024204090508
+      "required_token_positions_per_second": 9578.082105822976,
+      "required_pages_per_second_upper_estimate": 0.004010142964609463
     },
     {
       "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": 18959.742853197662,
-      "required_pages_per_second_upper_estimate": 0.003948814434278353
+      "required_token_positions_per_second": 19156.16421164595,
+      "required_pages_per_second_upper_estimate": 0.003989723823263571
     },
     {
       "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": 28439.614279796493,
-      "required_pages_per_second_upper_estimate": 0.0017049114512971706
+      "required_token_positions_per_second": 28734.24631746893,
+      "required_pages_per_second_upper_estimate": 0.0017225741920785605
     },
     {
       "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": 28439.614279796493,
-      "required_pages_per_second_upper_estimate": 0.0022110549176111586
+      "required_token_positions_per_second": 28734.24631746893,
+      "required_pages_per_second_upper_estimate": 0.0022339612625908167
     },
     {
       "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": 28439.614279796493,
-      "required_pages_per_second_upper_estimate": 0.3647928359025217
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28734.24631746893,
+      "required_pages_per_second_upper_estimate": 0.368572059330549
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.12564865240602785,
-    "producer_p95_seconds": 0.2857425729998795,
-    "gpu_update_seconds": 0.9721650838155405,
-    "producer_batches_per_second": 7.958700557874238,
-    "gpu_batches_per_second": 1.0286318822264358,
-    "producer_headroom": 7.737170794908597,
+    "producer_batch_seconds": 0.48308491897651606,
+    "producer_p95_seconds": 0.27002714300033404,
+    "gpu_update_seconds": 0.9621968049738424,
+    "producer_batches_per_second": 2.0700294310959695,
+    "gpu_batches_per_second": 1.0392884229408612,
+    "producer_headroom": 1.9917757048023623,
     "sustainable": true,
-    "buffer_seconds": 60.27423519656351,
+    "buffer_seconds": 60.61839871335207,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326950.939380991,
+    "generation_bytes_per_second": 290969.72877663944,
     "upload_bytes_per_second": 5293873.765262302,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.15606056282327,
+    "estimated_full_shards_per_hour": 62.435330366844056,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 626200, "loss": 0.5858222246170044, "lr": 0.0001, "grad_norm": 0.16567491129721723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91659315953, "tokens_per_second": 152543.8206018042, "data_wait_s": 1.3728000340051949e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626300, "loss": 0.5707839727401733, "lr": 0.0001, "grad_norm": 0.16334502314435753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91674039090, "tokens_per_second": 151943.28471567607, "data_wait_s": 1.7668997315922752e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626400, "loss": 0.5396880507469177, "lr": 0.0001, "grad_norm": 0.14936795053782706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91688532080, "tokens_per_second": 150298.5900370969, "data_wait_s": 1.7931997717823833e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 626500, "loss": 0.5421844124794006, "lr": 0.0001, "grad_norm": 0.16092432121913808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91703019884, "tokens_per_second": 138728.96300380188, "data_wait_s": 1.2681000953307375e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a0f8510542f25b2f68ae03eef5449400fbdaf016cebc7401bee2caad9859b8d
-size 470040
+oid sha256:67fc25c3b8df3021c728bf15d18c5018b8d6620792afbd6f459d71b653bcebde
+size 472158