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": 77115.46307926507,
-      "required_pages_per_second_upper_estimate": 0.005335026668349993
+      "required_token_positions_per_second": 76796.9126483558,
+      "required_pages_per_second_upper_estimate": 0.005312988610400857
     },
     {
       "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": 9639.432884908134,
-      "required_pages_per_second_upper_estimate": 0.004035829254662467
+      "required_token_positions_per_second": 9599.614081044476,
+      "required_pages_per_second_upper_estimate": 0.004019157952995932
     },
     {
       "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": 19278.865769816268,
-      "required_pages_per_second_upper_estimate": 0.004015279321972757
+      "required_token_positions_per_second": 19199.22816208895,
+      "required_pages_per_second_upper_estimate": 0.0039986929084681044
     },
     {
       "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": 28918.2986547244,
-      "required_pages_per_second_upper_estimate": 0.0017336078486653791
+      "required_token_positions_per_second": 28798.84224313343,
+      "required_pages_per_second_upper_estimate": 0.0017264466191898797
     },
     {
       "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": 28918.2986547244,
-      "required_pages_per_second_upper_estimate": 0.002248270522251755
+      "required_token_positions_per_second": 28798.84224313343,
+      "required_pages_per_second_upper_estimate": 0.0022389833116837825
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28918.2986547244,
-      "required_pages_per_second_upper_estimate": 0.37093288509285927
+      "required_token_positions_per_second": 28798.84224313343,
+      "required_pages_per_second_upper_estimate": 0.36940062650727196
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08643403580435916,
-    "producer_p95_seconds": 0.30955520499992417,
-    "gpu_update_seconds": 0.9560728426699172,
-    "producer_batches_per_second": 11.569516460662209,
-    "gpu_batches_per_second": 1.0459454085186777,
-    "producer_headroom": 11.061300490861719,
+    "producer_batch_seconds": 0.11026837050775384,
+    "producer_p95_seconds": 0.2940097300015623,
+    "gpu_update_seconds": 0.9600385934470047,
+    "producer_batches_per_second": 9.068783690148774,
+    "gpu_batches_per_second": 1.0416247917799997,
+    "producer_headroom": 8.706382338165565,
     "sustainable": true,
-    "buffer_seconds": 60.23258908820478,
+    "buffer_seconds": 60.4824313871613,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 495967.0083134804,
-    "upload_bytes_per_second": 4806584.188183917,
+    "generation_bytes_per_second": 517310.5303765397,
+    "upload_bytes_per_second": 4826287.9893601,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.42297446301755,
+    "estimated_full_shards_per_hour": 111.00279744598525,
     "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": 1.6209707380003238,
-      "bytes": 2622297,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.044743833001121,
       "bytes": 6332578,
@@ -562,6 +556,12 @@
       "bytes": 13424823,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.9566230939963134,
+      "bytes": 1183559,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 918800, "loss": 0.543188214302063, "lr": 3e-05, "grad_norm": 0.3316004527856014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134485361755, "tokens_per_second": 152932.4618170134, "data_wait_s": 1.8901002476923168e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 918900, "loss": 0.5481695532798767, "lr": 3e-05, "grad_norm": 0.34527990732742314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134500086125, "tokens_per_second": 153570.02215827053, "data_wait_s": 1.9810999219771475e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 919000, "loss": 0.5238047242164612, "lr": 3e-05, "grad_norm": 0.32704715253105254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134514824276, "tokens_per_second": 153763.470436366, "data_wait_s": 1.1886993888765574e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 919100, "loss": 0.5357107520103455, "lr": 3e-05, "grad_norm": 0.323277413238614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134529618891, "tokens_per_second": 155126.12917532533, "data_wait_s": 1.1288997484371066e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17650dfc5ac1ece9c2580b45f5479d6b9bd33616fb445c0b14a6f526e1c5a779
-size 724983
+oid sha256:60e63f4845211a8e5b8854eae5974bc913bbaf98d829c9e5366b5cb620cc8521
+size 727101