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": 77628.34084913018,
-      "required_pages_per_second_upper_estimate": 0.005370508742509643
+      "required_token_positions_per_second": 77593.92530322901,
+      "required_pages_per_second_upper_estimate": 0.005368127795189132
     },
     {
       "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": 9703.542606141273,
-      "required_pages_per_second_upper_estimate": 0.004062670656179566
+      "required_token_positions_per_second": 9699.240662903627,
+      "required_pages_per_second_upper_estimate": 0.004060869522380755
     },
     {
       "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": 19407.085212282545,
-      "required_pages_per_second_upper_estimate": 0.004041984050464397
+      "required_token_positions_per_second": 19398.481325807254,
+      "required_pages_per_second_upper_estimate": 0.004040192087811339
     },
     {
       "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": 29110.62781842382,
-      "required_pages_per_second_upper_estimate": 0.0017451376883595305
+      "required_token_positions_per_second": 29097.721988710877,
+      "required_pages_per_second_upper_estimate": 0.0017443640035743003
     },
     {
       "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": 29110.62781842382,
-      "required_pages_per_second_upper_estimate": 0.0022632232687627984
+      "required_token_positions_per_second": 29097.721988710877,
+      "required_pages_per_second_upper_estimate": 0.0022622198972693628
     },
     {
       "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": 29110.62781842382,
-      "required_pages_per_second_upper_estimate": 0.3733998770978287
+      "required_token_positions_per_second": 29097.721988710877,
+      "required_pages_per_second_upper_estimate": 0.37323433497147135
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10562178634353359,
-    "producer_p95_seconds": 0.3056859849966713,
-    "gpu_update_seconds": 0.9497562255425444,
-    "producer_batches_per_second": 9.467743678823156,
-    "gpu_batches_per_second": 1.052901758478871,
-    "producer_headroom": 8.992048500803364,
+    "producer_batch_seconds": 0.1180587853441466,
+    "producer_p95_seconds": 0.3112325040056021,
+    "gpu_update_seconds": 0.9501774747427537,
+    "producer_batches_per_second": 8.470356501508597,
+    "gpu_batches_per_second": 1.0524349677629803,
+    "producer_headroom": 8.048341950774304,
     "sustainable": true,
-    "buffer_seconds": 60.78439843472284,
+    "buffer_seconds": 60.81135838353624,
     "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": 253777.5153190255,
-    "upload_bytes_per_second": 5438731.173210582,
+    "generation_bytes_per_second": 271070.4629599066,
+    "upload_bytes_per_second": 5436961.105183009,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.45474714925836,
+    "estimated_full_shards_per_hour": 58.1654111537733,
     "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": 1.296304608000355,
-      "bytes": 7362354,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2885511929998756,
       "bytes": 10359776,
@@ -562,6 +556,12 @@
       "bytes": 10952650,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.072258756001247,
+      "bytes": 6000592,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 523300, "loss": 0.5911731123924255, "lr": 0.0003, "grad_norm": 0.1815925787559326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76593048711, "tokens_per_second": 154341.8191260696, "data_wait_s": 1.0716001270338893e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523400, "loss": 0.5877964496612549, "lr": 0.0003, "grad_norm": 0.18020763398318995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76607715438, "tokens_per_second": 153205.58404012618, "data_wait_s": 1.5669000276830047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 523500, "loss": 0.6117322444915771, "lr": 0.0003, "grad_norm": 0.15847204102332504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76622393886, "tokens_per_second": 153661.93508374627, "data_wait_s": 1.0378003935329616e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 523600, "loss": 0.5748156905174255, "lr": 0.0003, "grad_norm": 0.15414128055515142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76637125565, "tokens_per_second": 153879.30634632974, "data_wait_s": 1.1898002412635833e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15cf4fb2b82b532a4feb683a26b358a82ac8bc6d54ffdb9d43bac373ae789b0f
-size 821076
+oid sha256:f5da005b8f3641dbf58cdd496c8c18863b8f68cf70e0528f5aa7f2403705942b
+size 823194