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": 75468.40702326216,
-      "required_pages_per_second_upper_estimate": 0.006541410247847466
+      "required_token_positions_per_second": 75097.76905987824,
+      "required_pages_per_second_upper_estimate": 0.006509284288554677
     },
     {
       "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": 9433.55087790777,
-      "required_pages_per_second_upper_estimate": 0.0037631844893520703
+      "required_token_positions_per_second": 9387.22113248478,
+      "required_pages_per_second_upper_estimate": 0.0037447028612114168
     },
     {
       "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": 18867.10175581554,
-      "required_pages_per_second_upper_estimate": 0.0043645719415579265
+      "required_token_positions_per_second": 18774.44226496956,
+      "required_pages_per_second_upper_estimate": 0.004343136799102051
     },
     {
       "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": 28300.65263372331,
-      "required_pages_per_second_upper_estimate": 0.001723361263344678
+      "required_token_positions_per_second": 28161.66339745434,
+      "required_pages_per_second_upper_estimate": 0.0017148975480760632
     },
     {
       "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": 28300.65263372331,
-      "required_pages_per_second_upper_estimate": 0.002178662580964572
+      "required_token_positions_per_second": 28161.66339745434,
+      "required_pages_per_second_upper_estimate": 0.0021679628048097556
     },
     {
       "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": 28300.65263372331,
-      "required_pages_per_second_upper_estimate": 0.3513993895193925
+      "required_token_positions_per_second": 28161.66339745434,
+      "required_pages_per_second_upper_estimate": 0.3496736083719823
     }
   ],
   "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": 0.11611132431247029,
-    "producer_p95_seconds": 0.33163757100192015,
-    "gpu_update_seconds": 0.9769386013047061,
-    "producer_batches_per_second": 8.612424377392108,
-    "gpu_batches_per_second": 1.0236057810229786,
-    "producer_headroom": 8.413809825092,
+    "producer_batch_seconds": 0.10586516326557671,
+    "producer_p95_seconds": 0.31800649800061365,
+    "gpu_update_seconds": 0.9817601897229987,
+    "producer_batches_per_second": 9.44597797002748,
+    "gpu_batches_per_second": 1.018578681910241,
+    "producer_headroom": 9.273685123973445,
     "sustainable": true,
-    "buffer_seconds": 60.57019328089178,
+    "buffer_seconds": 60.86913176282592,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252701.605807888,
-    "upload_bytes_per_second": 2886964.7441037465,
+    "generation_bytes_per_second": 266863.4354399473,
+    "upload_bytes_per_second": 2848091.342899169,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.22388201405983,
+    "estimated_full_shards_per_hour": 57.26268098257841,
     "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": 1.1309446140003274,
-      "bytes": 12087341,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 2.7144779629998084,
       "bytes": 851325,
@@ -562,6 +556,12 @@
       "bytes": 2343640,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0508419269972364,
+      "bytes": 6132224,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -288,3 +288,4 @@
 {"step": 223800, "loss": 0.5782792679965496, "lr": 0.0003, "grad_norm": 0.1459301303786713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32757501567, "tokens_per_second": 149032.4429403084, "data_wait_s": 1.0058000043500215e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223900, "loss": 0.5707023125141859, "lr": 0.0003, "grad_norm": 0.15491452613289144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32772181501, "tokens_per_second": 148682.36673024838, "data_wait_s": 1.8599999748403206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 224000, "loss": 0.5858014151453972, "lr": 0.0003, "grad_norm": 0.15248701402295245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32786845177, "tokens_per_second": 148408.81860517786, "data_wait_s": 1.1183001333847642e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 224100, "loss": 0.561779797077179, "lr": 0.0003, "grad_norm": 0.1768840042951018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32801602590, "tokens_per_second": 150094.3845559986, "data_wait_s": 9.258998034056276e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59e3ae7bf4a8f9b7ac0c2cabb823e11db214fc97af0c09a50214d7689a820ede
-size 618300
+oid sha256:f9d2aa7ca9c57b01144d6d23a4119cb5f16da7ba6bf12f9630598dbb9628139b
+size 620418