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": 75577.23972103799,
-      "required_pages_per_second_upper_estimate": 0.0065508435902567235
+      "required_token_positions_per_second": 76054.68664368195,
+      "required_pages_per_second_upper_estimate": 0.00659222747943334
     },
     {
       "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": 9447.154965129748,
-      "required_pages_per_second_upper_estimate": 0.0037686113631441474
+      "required_token_positions_per_second": 9506.835830460244,
+      "required_pages_per_second_upper_estimate": 0.003792418952632936
     },
     {
       "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": 18894.309930259496,
-      "required_pages_per_second_upper_estimate": 0.00437086607386802
+      "required_token_positions_per_second": 19013.67166092049,
+      "required_pages_per_second_upper_estimate": 0.004398478309561729
     },
     {
       "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": 28341.464895389243,
-      "required_pages_per_second_upper_estimate": 0.0017258465159547447
+      "required_token_positions_per_second": 28520.507491380737,
+      "required_pages_per_second_upper_estimate": 0.0017367492706867234
     },
     {
       "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": 28341.464895389243,
-      "required_pages_per_second_upper_estimate": 0.002181804421843186
+      "required_token_positions_per_second": 28520.507491380737,
+      "required_pages_per_second_upper_estimate": 0.0021955876165042384
     },
     {
       "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": 28341.464895389243,
-      "required_pages_per_second_upper_estimate": 0.3519061412194301
+      "required_token_positions_per_second": 28520.507491380737,
+      "required_pages_per_second_upper_estimate": 0.3541292510446222
     }
   ],
   "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": 0.24644996732058644,
-    "producer_p95_seconds": 0.3844791110022925,
-    "gpu_update_seconds": 0.9755317906837603,
-    "producer_batches_per_second": 4.057618716172043,
-    "gpu_batches_per_second": 1.0250819189593912,
-    "producer_headroom": 3.958336052099254,
+    "producer_batch_seconds": 0.11060982517977891,
+    "producer_p95_seconds": 0.3755583900056081,
+    "gpu_update_seconds": 0.9694077150750414,
+    "producer_batches_per_second": 9.040788179301947,
+    "gpu_batches_per_second": 1.0315577072982036,
+    "producer_headroom": 8.764209811374544,
     "sustainable": true,
-    "buffer_seconds": 60.48297102239314,
+    "buffer_seconds": 60.103278334652565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 562044.1916043733,
-    "upload_bytes_per_second": 5401733.275667653,
+    "generation_bytes_per_second": 574102.0126152907,
+    "upload_bytes_per_second": 5552254.925416074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.30079989957046,
+    "estimated_full_shards_per_hour": 61.59446374818821,
     "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": 2.272034231998987,
-      "bytes": 5974861,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2107169869996142,
       "bytes": 7639876,
@@ -562,6 +556,12 @@
       "bytes": 6179585,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3004778859976796,
+      "bytes": 12330619,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 73400, "loss": 0.6100993007421494, "lr": 0.0003, "grad_norm": 0.12337211974140161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10741768291, "tokens_per_second": 146735.5201714556, "data_wait_s": 1.044199598254636e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 73500, "loss": 0.6147768199443817, "lr": 0.0003, "grad_norm": 0.12840102069003204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10756446645, "tokens_per_second": 150517.09152512488, "data_wait_s": 1.0752999514807016e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 73600, "loss": 0.6206010710448027, "lr": 0.0003, "grad_norm": 0.1478000530908829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10771041359, "tokens_per_second": 149144.17894786637, "data_wait_s": 9.26100619835779e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 73700, "loss": 0.6197205875068903, "lr": 0.0003, "grad_norm": 0.13492088232152857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10785605218, "tokens_per_second": 148931.69318447958, "data_wait_s": 8.645998605061322e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:967789985bfcd9b3bdb421b2d0735c3979e621dff205ab3c1c80bde205cb1cc4
-size 716511
+oid sha256:7aa4f4c2297c050ad5f56cd042ccc6de51389acb1e022275eb0fbe4cf4201fd0
+size 718629