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": 74984.8688873846,
-      "required_pages_per_second_upper_estimate": 0.006499498387745799
+      "required_token_positions_per_second": 75696.60753333535,
+      "required_pages_per_second_upper_estimate": 0.006561190089691705
     },
     {
       "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": 9373.108610923075,
-      "required_pages_per_second_upper_estimate": 0.003739073165359452
+      "required_token_positions_per_second": 9462.075941666919,
+      "required_pages_per_second_upper_estimate": 0.003774563563773304
     },
     {
       "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": 18746.21722184615,
-      "required_pages_per_second_upper_estimate": 0.004336607432119243
+      "required_token_positions_per_second": 18924.151883333838,
+      "required_pages_per_second_upper_estimate": 0.004377769484511333
     },
     {
       "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": 28119.325832769224,
-      "required_pages_per_second_upper_estimate": 0.0017123194125147744
+      "required_token_positions_per_second": 28386.22782500076,
+      "required_pages_per_second_upper_estimate": 0.0017285723435151448
     },
     {
       "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": 28119.325832769224,
-      "required_pages_per_second_upper_estimate": 0.0021647035418824153
+      "required_token_positions_per_second": 28386.22782500076,
+      "required_pages_per_second_upper_estimate": 0.0021852503960764005
     },
     {
       "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": 28119.325832769224,
-      "required_pages_per_second_upper_estimate": 0.3491479175133072
+      "required_token_positions_per_second": 28386.22782500076,
+      "required_pages_per_second_upper_estimate": 0.35246194699331684
     }
   ],
   "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": 0.25891980108596613,
-    "producer_p95_seconds": 0.31517380600053,
-    "gpu_update_seconds": 0.9832383665393586,
-    "producer_batches_per_second": 3.86219978466607,
-    "gpu_batches_per_second": 1.0170473753171738,
-    "producer_headroom": 3.797463007523729,
+    "producer_batch_seconds": 0.18506221920318922,
+    "producer_p95_seconds": 0.2764886689983541,
+    "gpu_update_seconds": 0.973993450994902,
+    "producer_batches_per_second": 5.403588070572358,
+    "gpu_batches_per_second": 1.0267009485315668,
+    "producer_headroom": 5.263059392611655,
     "sustainable": true,
-    "buffer_seconds": 60.960778725440235,
+    "buffer_seconds": 60.38759396168392,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 588830.115475454,
-    "upload_bytes_per_second": 5681621.317287129,
+    "generation_bytes_per_second": 568058.7247640388,
+    "upload_bytes_per_second": 5651517.664839139,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.17461775874002,
+    "estimated_full_shards_per_hour": 60.946089302019466,
     "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": 1.289128628999606,
-      "bytes": 10630071,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0585477320000791,
       "bytes": 5855284,
@@ -562,6 +556,12 @@
       "bytes": 7061925,
       "pages": 24,
       "shards": 3
+    },
+    {
+      "seconds": 1.0121167409997724,
+      "bytes": 6826316,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 135000, "loss": 0.6044514365494251, "lr": 0.0003, "grad_norm": 0.12881867767854382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19757015197, "tokens_per_second": 149972.18968166088, "data_wait_s": 8.879997039912269e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 135100, "loss": 0.5784531645476818, "lr": 0.0003, "grad_norm": 0.12849210119629034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19771706896, "tokens_per_second": 133653.35975033976, "data_wait_s": 1.019000046653673e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 135200, "loss": 0.620394753292203, "lr": 0.0003, "grad_norm": 0.14046893809918987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19786376990, "tokens_per_second": 149761.500091097, "data_wait_s": 2.079499972751364e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 135300, "loss": 0.6356552727520466, "lr": 0.0003, "grad_norm": 0.15478166533480578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19800839084, "tokens_per_second": 146629.99899440585, "data_wait_s": 9.978997695725411e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41a6541b30dd1259b2a3df245b7d0fa0c291e95604758432815d4508d3ec3f63
-size 431133
+oid sha256:fa7d985558b7dbd29c268c1fe3291817c8ccf2494ccefb671aeb013a13c6ce21
+size 433251