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": 75662.00814788805,
-      "required_pages_per_second_upper_estimate": 0.006558191102652473
+      "required_token_positions_per_second": 75199.18650291045,
+      "required_pages_per_second_upper_estimate": 0.00651807489547655
     },
     {
       "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": 9457.751018486006,
-      "required_pages_per_second_upper_estimate": 0.003772838287253074
+      "required_token_positions_per_second": 9399.898312863806,
+      "required_pages_per_second_upper_estimate": 0.0037497599780053482
     },
     {
       "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": 18915.50203697201,
-      "required_pages_per_second_upper_estimate": 0.004375768494787621
+      "required_token_positions_per_second": 18799.796625727613,
+      "required_pages_per_second_upper_estimate": 0.0043490020842419175
     },
     {
       "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": 28373.25305545802,
-      "required_pages_per_second_upper_estimate": 0.0017277822481233467
+      "required_token_positions_per_second": 28199.69493859142,
+      "required_pages_per_second_upper_estimate": 0.001717213469395236
     },
     {
       "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": 28373.25305545802,
-      "required_pages_per_second_upper_estimate": 0.0021842515624005397
+      "required_token_positions_per_second": 28199.69493859142,
+      "required_pages_per_second_upper_estimate": 0.0021708905781245317
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28373.25305545802,
-      "required_pages_per_second_upper_estimate": 0.35230084377935633
+      "required_token_positions_per_second": 28199.69493859142,
+      "required_pages_per_second_upper_estimate": 0.35014583282952455
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11217890628907412,
-    "producer_p95_seconds": 0.2792825359974813,
-    "gpu_update_seconds": 0.9744388472467205,
-    "producer_batches_per_second": 8.914331874685045,
-    "gpu_batches_per_second": 1.0262316643322489,
-    "producer_headroom": 8.686471275942791,
+    "producer_batch_seconds": 0.1807987464920302,
+    "producer_p95_seconds": 0.2982292699998652,
+    "gpu_update_seconds": 0.9804361380577765,
+    "producer_batches_per_second": 5.531011798492093,
+    "gpu_batches_per_second": 1.0199542440173401,
+    "producer_headroom": 5.422803847265585,
     "sustainable": true,
-    "buffer_seconds": 60.41520852929667,
+    "buffer_seconds": 60.787040559582145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377707.48333928525,
-    "upload_bytes_per_second": 6450756.5794556225,
+    "generation_bytes_per_second": 398140.9361914577,
+    "upload_bytes_per_second": 6506501.200829124,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.04723334440153,
+    "estimated_full_shards_per_hour": 85.43177666003987,
     "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": 1.1366644360005012,
-      "bytes": 5837515,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.8290303609992407,
       "bytes": 5336251,
@@ -562,6 +556,12 @@
       "bytes": 10590685,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9887308050019783,
+      "bytes": 8510426,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 60600, "loss": 0.596485260874033, "lr": 0.0003, "grad_norm": 0.12365103573619238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8869240999, "tokens_per_second": 148779.7304257529, "data_wait_s": 1.3098000636091456e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60700, "loss": 0.6107595860958099, "lr": 0.0003, "grad_norm": 0.14709025953583868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8884063213, "tokens_per_second": 151259.12823683946, "data_wait_s": 1.0274001397192478e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 60800, "loss": 0.6119226086884737, "lr": 0.0003, "grad_norm": 0.13236256579612282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8898722185, "tokens_per_second": 149523.22127715946, "data_wait_s": 7.384998752968386e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 60900, "loss": 0.5926342438906431, "lr": 0.0003, "grad_norm": 0.1305635818069234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8913392632, "tokens_per_second": 149943.4503468802, "data_wait_s": 8.844999683788046e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7e94f07094da77e237198ca79840516471e30f56cb1924387f2503a76841f1c
-size 443841
+oid sha256:6969fa1b634972eb5bb02cd6b6b333b6a537afb9f76efb748214ce933ce225db
+size 445959