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": 74784.17727081999,
-      "required_pages_per_second_upper_estimate": 0.006482102947070231
+      "required_token_positions_per_second": 74769.53455800051,
+      "required_pages_per_second_upper_estimate": 0.006480833753834655
     },
     {
       "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": 9348.022158852498,
-      "required_pages_per_second_upper_estimate": 0.003729065804552616
+      "required_token_positions_per_second": 9346.191819750064,
+      "required_pages_per_second_upper_estimate": 0.003728335654918647
     },
     {
       "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": 18696.044317704997,
-      "required_pages_per_second_upper_estimate": 0.0043250008137591415
+      "required_token_positions_per_second": 18692.383639500127,
+      "required_pages_per_second_upper_estimate": 0.004324153980282181
     },
     {
       "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": 28044.066476557495,
-      "required_pages_per_second_upper_estimate": 0.0017077365125768061
+      "required_token_positions_per_second": 28038.57545925019,
+      "required_pages_per_second_upper_estimate": 0.0017074021384319328
     },
     {
       "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": 28044.066476557495,
-      "required_pages_per_second_upper_estimate": 0.0021589098683100047
+      "required_token_positions_per_second": 28038.57545925019,
+      "required_pages_per_second_upper_estimate": 0.0021584871545975797
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28044.066476557495,
-      "required_pages_per_second_upper_estimate": 0.34821344818602
+      "required_token_positions_per_second": 28038.57545925019,
+      "required_pages_per_second_upper_estimate": 0.3481452681283161
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0417184604140628,
-    "producer_p95_seconds": 0.12838051699998232,
-    "gpu_update_seconds": 0.9858769955174449,
-    "producer_batches_per_second": 23.970203839615134,
-    "gpu_batches_per_second": 1.0143253210560437,
-    "producer_headroom": 23.63167254334049,
+    "producer_batch_seconds": 0.3080829623515635,
+    "producer_p95_seconds": 0.13464080500000364,
+    "gpu_update_seconds": 0.986070067653122,
+    "producer_batches_per_second": 3.2458789423702936,
+    "gpu_batches_per_second": 1.0141267165527412,
+    "producer_headroom": 3.2006640682969194,
     "sustainable": true,
-    "buffer_seconds": 60.13849672656414,
+    "buffer_seconds": 60.15027412684044,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 41,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358180.446823969,
+    "generation_bytes_per_second": 320415.8281157413,
     "upload_bytes_per_second": 420340.40760150086,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.85718587436011,
+    "estimated_full_shards_per_hour": 68.75377781490498,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 315100, "loss": 0.5646674651652575, "lr": 0.0003, "grad_norm": 0.17352291347570195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46121906842, "tokens_per_second": 46362.37927092308, "data_wait_s": 7.901999993009667e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 315200, "loss": 0.5742356237024069, "lr": 0.0003, "grad_norm": 0.17197717698161305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46136530075, "tokens_per_second": 148760.2792835282, "data_wait_s": 1.0716000019783678e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 315300, "loss": 0.6177494637668133, "lr": 0.0003, "grad_norm": 0.17060833492964608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46151259292, "tokens_per_second": 149500.7609276313, "data_wait_s": 1.2448000006770599e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 315400, "loss": 0.6212431937456131, "lr": 0.0003, "grad_norm": 0.1611149199047263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46165932689, "tokens_per_second": 149151.70759550997, "data_wait_s": 1.2699999842880061e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d10532281bddc4a69817f2ab9d4228b53d99009bf6497815a78261b4963f0f5
-size 6519
+oid sha256:1dbefdf64bedd18b35c97f910981012fd0d15b3deb9aa46254348438aac5a6a5
+size 8637