Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75343.2066073082,
-      "required_pages_per_second_upper_estimate": 0.00521241785341845
+      "required_token_positions_per_second": 75500.7184057323,
+      "required_pages_per_second_upper_estimate": 0.005223314885111155
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9417.900825913524,
-      "required_pages_per_second_upper_estimate": 0.003943078407676869
+      "required_token_positions_per_second": 9437.589800716538,
+      "required_pages_per_second_upper_estimate": 0.003951321770274342
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18835.80165182705,
-      "required_pages_per_second_upper_estimate": 0.003923000750582135
+      "required_token_positions_per_second": 18875.179601433076,
+      "required_pages_per_second_upper_estimate": 0.003931202139018705
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28253.70247774057,
-      "required_pages_per_second_upper_estimate": 0.0016937663226348032
+      "required_token_positions_per_second": 28312.76940214961,
+      "required_pages_per_second_upper_estimate": 0.0016973072945631537
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28253.70247774057,
-      "required_pages_per_second_upper_estimate": 0.0021966010927409064
+      "required_token_positions_per_second": 28312.76940214961,
+      "required_pages_per_second_upper_estimate": 0.002201193286305766
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28253.70247774057,
-      "required_pages_per_second_upper_estimate": 0.362408158922289
+      "required_token_positions_per_second": 28312.76940214961,
+      "required_pages_per_second_upper_estimate": 0.36316580600748594
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06219016758592666,
-    "producer_p95_seconds": 0.1501125699999193,
-    "gpu_update_seconds": 0.9785620140150563,
-    "producer_batches_per_second": 16.07971225705935,
-    "gpu_batches_per_second": 1.0219076417006863,
-    "producer_headroom": 15.734995611050584,
+    "producer_batch_seconds": 0.09342547477342755,
+    "producer_p95_seconds": 0.15122849099998348,
+    "gpu_update_seconds": 0.9765205094313685,
+    "producer_batches_per_second": 10.703718685135588,
+    "gpu_batches_per_second": 1.0240440321958049,
+    "producer_headroom": 10.452400823218662,
     "sustainable": true,
-    "buffer_seconds": 60.670844868933486,
+    "buffer_seconds": 60.54427158474485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401273.3133146612,
-    "upload_bytes_per_second": 3639884.432836335,
+    "generation_bytes_per_second": 420539.77038652933,
+    "upload_bytes_per_second": 3544522.0528646167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.10391187267186,
+    "estimated_full_shards_per_hour": 90.2380450601283,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 10320081,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3121002539999154,
+      "bytes": 2619780,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 843900, "loss": 0.5337322354316711, "lr": 3e-05, "grad_norm": 0.30476746880164146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123523342733, "tokens_per_second": 149110.35123163465, "data_wait_s": 1.6943999980867375e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844000, "loss": 0.5025510191917419, "lr": 3e-05, "grad_norm": 0.2940395860061583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123537879030, "tokens_per_second": 150556.11786133875, "data_wait_s": 1.715500002319459e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 844100, "loss": 0.5470116138458252, "lr": 3e-05, "grad_norm": 0.29249404624772823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123552492254, "tokens_per_second": 150459.95035574777, "data_wait_s": 1.5677999954277766e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 844200, "loss": 0.5173737406730652, "lr": 3e-05, "grad_norm": 0.2874194802671103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123567140903, "tokens_per_second": 151268.59734326255, "data_wait_s": 1.1069999800383812e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be23a447dbb248900dbc04dd1c241ffea64f854063d801538a2e4042b832ff96
-size 34053
+oid sha256:37de9e7f0a260a47d39d9c36bb49d824e7a9c82f7b06ce19c18c5488337dbf58
+size 36171