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": 75535.3813282497,
-      "required_pages_per_second_upper_estimate": 0.006547215410858962
+      "required_token_positions_per_second": 74733.86792011026,
+      "required_pages_per_second_upper_estimate": 0.006477742260058596
     },
     {
       "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": 9441.922666031212,
-      "required_pages_per_second_upper_estimate": 0.0037665241208038984
+      "required_token_positions_per_second": 9341.733490013783,
+      "required_pages_per_second_upper_estimate": 0.0037265571605288746
     },
     {
       "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": 18883.845332062425,
-      "required_pages_per_second_upper_estimate": 0.004368445273245766
+      "required_token_positions_per_second": 18683.466980027566,
+      "required_pages_per_second_upper_estimate": 0.004322091268041051
     },
     {
       "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": 28325.767998093637,
-      "required_pages_per_second_upper_estimate": 0.0017248906572646977
+      "required_token_positions_per_second": 28025.200470041353,
+      "required_pages_per_second_upper_estimate": 0.0017065876717622596
     },
     {
       "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": 28325.767998093637,
-      "required_pages_per_second_upper_estimate": 0.002180596031237578
+      "required_token_positions_per_second": 28025.200470041353,
+      "required_pages_per_second_upper_estimate": 0.0021574575109039435
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28325.767998093637,
-      "required_pages_per_second_upper_estimate": 0.3517112382891545
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28025.200470041353,
+      "required_pages_per_second_upper_estimate": 0.3479791955255516
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11099471561712448,
-    "producer_p95_seconds": 0.3792780860021594,
-    "gpu_update_seconds": 0.9760723875822448,
-    "producer_batches_per_second": 9.009437921797044,
-    "gpu_batches_per_second": 1.0245141781717897,
-    "producer_headroom": 8.793863583102459,
+    "producer_batch_seconds": 0.2247381723514934,
+    "producer_p95_seconds": 0.394886788999429,
+    "gpu_update_seconds": 0.9865406682658854,
+    "producer_batches_per_second": 4.449622374057519,
+    "gpu_batches_per_second": 1.0136429568157317,
+    "producer_headroom": 4.38973343043354,
     "sustainable": true,
-    "buffer_seconds": 60.516488030099175,
+    "buffer_seconds": 60.178980764219006,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297295.53941490006,
-    "upload_bytes_per_second": 5315379.803461301,
+    "generation_bytes_per_second": 311246.3765358246,
+    "upload_bytes_per_second": 5405043.676841065,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.79270207247974,
+    "estimated_full_shards_per_hour": 66.78622695976308,
     "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": 1.3565457900003821,
-      "bytes": 7973344,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.262600800000655,
       "bytes": 11885196,
@@ -562,6 +556,12 @@
       "bytes": 2645607,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5463334099986241,
+      "bytes": 15909077,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 226500, "loss": 0.6257619615644217, "lr": 0.0003, "grad_norm": 0.15677036568711947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33153327328, "tokens_per_second": 148919.01146690917, "data_wait_s": 9.450999641558155e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 226600, "loss": 0.5676675904542208, "lr": 0.0003, "grad_norm": 0.13611202505926445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33168023130, "tokens_per_second": 148936.78350129872, "data_wait_s": 8.999999408842996e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 226700, "loss": 0.6044607497751713, "lr": 0.0003, "grad_norm": 0.14674524639210093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33182695458, "tokens_per_second": 148143.8770490476, "data_wait_s": 1.906300167320296e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 226800, "loss": 0.5968139693140984, "lr": 0.0003, "grad_norm": 0.13900636810270997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33197378430, "tokens_per_second": 150224.2463812641, "data_wait_s": 1.4385001122718677e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c105fc0e21f56d0363e505a5d1c515ea1eef6652841cc41d1d17a57e6a6a7ac3
-size 676269
+oid sha256:efadc1fb3c4159f65ed96353882cedd7cf6d0db7b47411eb7ec679b9b902be22
+size 678387