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": 75348.81776177572,
-      "required_pages_per_second_upper_estimate": 0.005212806046104186
+      "required_token_positions_per_second": 76742.60649064253,
+      "required_pages_per_second_upper_estimate": 0.0053092315844026055
     },
     {
       "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": 9418.602220221965,
-      "required_pages_per_second_upper_estimate": 0.003943372066827034
+      "required_token_positions_per_second": 9592.825811330316,
+      "required_pages_per_second_upper_estimate": 0.004016315846221805
     },
     {
       "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": 18837.20444044393,
-      "required_pages_per_second_upper_estimate": 0.003923292914457008
+      "required_token_positions_per_second": 19185.651622660633,
+      "required_pages_per_second_upper_estimate": 0.003995865273342607
     },
     {
       "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": 28255.806660665894,
-      "required_pages_per_second_upper_estimate": 0.001693892465188275
+      "required_token_positions_per_second": 28778.47743399095,
+      "required_pages_per_second_upper_estimate": 0.001725225779977051
     },
     {
       "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": 28255.806660665894,
-      "required_pages_per_second_upper_estimate": 0.0021967646837080276
+      "required_token_positions_per_second": 28778.47743399095,
+      "required_pages_per_second_upper_estimate": 0.0022374000373482834
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28255.806660665894,
-      "required_pages_per_second_upper_estimate": 0.36243514912155944
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28778.47743399095,
+      "required_pages_per_second_upper_estimate": 0.36913940860162064
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07565455060171189,
-    "producer_p95_seconds": 0.27016047299912316,
-    "gpu_update_seconds": 0.9784891414368289,
-    "producer_batches_per_second": 13.217975548682624,
-    "gpu_batches_per_second": 1.0219837478539457,
-    "producer_headroom": 12.933645546163456,
+    "producer_batch_seconds": 0.13259479776556304,
+    "producer_p95_seconds": 0.27210508300049696,
+    "gpu_update_seconds": 0.9607179554031682,
+    "producer_batches_per_second": 7.541774012643169,
+    "gpu_batches_per_second": 1.0408882173752514,
+    "producer_headroom": 7.2455177095392935,
     "sustainable": true,
-    "buffer_seconds": 60.66632676908339,
+    "buffer_seconds": 60.525231190399595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269078.7183893529,
-    "upload_bytes_per_second": 5324650.759120141,
+    "generation_bytes_per_second": 269166.84837642295,
+    "upload_bytes_per_second": 5291585.741498411,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.73802913437309,
+    "estimated_full_shards_per_hour": 57.75693977803723,
     "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": 1.6794259740017878,
-      "bytes": 11268683,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.0690768630011007,
       "bytes": 5982757,
@@ -562,6 +556,12 @@
       "bytes": 10114603,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2068859079990943,
+      "bytes": 6031296,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 748500, "loss": 0.517221212387085, "lr": 5e-05, "grad_norm": 0.2359657045641018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109559487637, "tokens_per_second": 149969.55566285597, "data_wait_s": 1.4731998817296699e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 748600, "loss": 0.514732837677002, "lr": 5e-05, "grad_norm": 0.24052960704157414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109574050392, "tokens_per_second": 150135.04202254055, "data_wait_s": 8.305996743729338e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 748700, "loss": 0.5358645915985107, "lr": 5e-05, "grad_norm": 0.23892753755651322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109588654043, "tokens_per_second": 150922.2132229246, "data_wait_s": 1.5794998034834862e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 748800, "loss": 0.5097352862358093, "lr": 5e-05, "grad_norm": 0.2574977276468063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109603179872, "tokens_per_second": 149288.6030433589, "data_wait_s": 1.4174001989886165e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b9e9e744899087ce2e2b73f0f347ef9aba3feadd754f017ac228155d9692ea9
-size 611946
+oid sha256:0ae3d36cf767a6319a9f4f36e5f2b42a4da49d01df0f9d1f7c4f9a3cecd16bbc
+size 614064