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": 76095.36622040457,
-      "required_pages_per_second_upper_estimate": 0.006595753482042212
+      "required_token_positions_per_second": 73907.42228458513,
+      "required_pages_per_second_upper_estimate": 0.006406108046978591
     },
     {
       "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": 9511.920777550571,
-      "required_pages_per_second_upper_estimate": 0.0037944474140540014
+      "required_token_positions_per_second": 9238.427785573142,
+      "required_pages_per_second_upper_estimate": 0.003685346970469579
     },
     {
       "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": 19023.841555101142,
-      "required_pages_per_second_upper_estimate": 0.004400830935596399
+      "required_token_positions_per_second": 18476.855571146283,
+      "required_pages_per_second_upper_estimate": 0.0042742953548329694
     },
     {
       "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": 28535.762332651713,
-      "required_pages_per_second_upper_estimate": 0.001737678210484159
+      "required_token_positions_per_second": 27715.283356719425,
+      "required_pages_per_second_upper_estimate": 0.001687715345570383
     },
     {
       "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": 28535.762332651713,
-      "required_pages_per_second_upper_estimate": 0.0021967619764133813
+      "required_token_positions_per_second": 27715.283356719425,
+      "required_pages_per_second_upper_estimate": 0.0021335992336149433
     },
     {
       "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": 28535.762332651713,
-      "required_pages_per_second_upper_estimate": 0.3543186651185382
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27715.283356719425,
+      "required_pages_per_second_upper_estimate": 0.3441310622039488
     }
   ],
   "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": 0.10179032871093341,
-    "producer_p95_seconds": 0.4072658369987039,
-    "gpu_update_seconds": 0.9688894825271269,
-    "producer_batches_per_second": 9.824116030117397,
-    "gpu_batches_per_second": 1.0321094593696367,
-    "producer_headroom": 9.518482696706897,
+    "producer_batch_seconds": 0.124362981648801,
+    "producer_p95_seconds": 0.40652090799994767,
+    "gpu_update_seconds": 0.997572337404838,
+    "producer_batches_per_second": 8.040978004403138,
+    "gpu_batches_per_second": 1.0024335704831968,
+    "producer_headroom": 8.021457222873329,
     "sustainable": true,
-    "buffer_seconds": 60.071147916681866,
+    "buffer_seconds": 60.85191258169512,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326417.2543513872,
-    "upload_bytes_per_second": 6182411.037728616,
+    "generation_bytes_per_second": 320619.61218343594,
+    "upload_bytes_per_second": 6292706.395301407,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.04154417902195,
+    "estimated_full_shards_per_hour": 68.79750513198194,
     "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": 0.7280326340005558,
-      "bytes": 1168675,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1539005899976473,
       "bytes": 5543356,
@@ -562,6 +556,12 @@
       "bytes": 900379,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.208673165005166,
+      "bytes": 11672731,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"step": 77900, "loss": 0.648516247048974, "lr": 0.0003, "grad_norm": 0.133658089222264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11400499219, "tokens_per_second": 149770.91388315082, "data_wait_s": 9.019000572152436e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 78000, "loss": 0.6717341057956219, "lr": 0.0003, "grad_norm": 0.1393901915487311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11415141762, "tokens_per_second": 149919.69667152414, "data_wait_s": 1.1878997611347586e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 78100, "loss": 0.6047646552324295, "lr": 0.0003, "grad_norm": 0.13520182350403326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11429812036, "tokens_per_second": 148336.8116493828, "data_wait_s": 1.0003001079894602e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 78200, "loss": 0.6694426946341991, "lr": 0.0003, "grad_norm": 0.14425529049851427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11444504104, "tokens_per_second": 150539.53375494335, "data_wait_s": 1.3578996004071087e-05, "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:cf1d9f57533383d5eda987a837ebc6f977f1e22b1aa422e50b68ef7f9a7a2350
-size 812604
+oid sha256:937bae56d83bd8818acdab93384e7c4c2aea79f1a6e43ec881cf838e39cb8b80
+size 814722