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": 76624.75053199734,
-      "required_pages_per_second_upper_estimate": 0.005301078035720032
+      "required_token_positions_per_second": 76818.21459025308,
+      "required_pages_per_second_upper_estimate": 0.005314462328168625
     },
     {
       "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": 9578.093816499668,
-      "required_pages_per_second_upper_estimate": 0.004010147867625247
+      "required_token_positions_per_second": 9602.276823781634,
+      "required_pages_per_second_upper_estimate": 0.004020272787775589
     },
     {
       "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": 19156.187632999336,
-      "required_pages_per_second_upper_estimate": 0.003989728701313818
+      "required_token_positions_per_second": 19204.55364756327,
+      "required_pages_per_second_upper_estimate": 0.003999802066649908
     },
     {
       "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": 28734.281449499005,
-      "required_pages_per_second_upper_estimate": 0.00172257629819012
+      "required_token_positions_per_second": 28806.830471344903,
+      "required_pages_per_second_upper_estimate": 0.0017269255012738393
     },
     {
       "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": 28734.281449499005,
-      "required_pages_per_second_upper_estimate": 0.0022339639939516257
+      "required_token_positions_per_second": 28806.830471344903,
+      "required_pages_per_second_upper_estimate": 0.002239604361290722
     },
     {
       "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": 28734.281449499005,
-      "required_pages_per_second_upper_estimate": 0.368572509966509
+      "required_token_positions_per_second": 28806.830471344903,
+      "required_pages_per_second_upper_estimate": 0.36950309092167755
     }
   ],
   "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": 0.08998299506249907,
-    "producer_p95_seconds": 0.214224072999059,
-    "gpu_update_seconds": 0.9621956285418808,
-    "producer_batches_per_second": 11.113210882849971,
-    "gpu_batches_per_second": 1.0392896936306064,
-    "producer_headroom": 10.6930829305423,
+    "producer_batch_seconds": 0.10314217333585418,
+    "producer_p95_seconds": 0.20737120300145762,
+    "gpu_update_seconds": 0.9597723716082673,
+    "producer_batches_per_second": 9.695355136097186,
+    "gpu_batches_per_second": 1.0419137178582503,
+    "producer_headroom": 9.305333992556392,
     "sustainable": true,
-    "buffer_seconds": 60.61832459813849,
+    "buffer_seconds": 60.46565941132084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351633.163037833,
-    "upload_bytes_per_second": 5253573.894710322,
+    "generation_bytes_per_second": 365888.4023332574,
+    "upload_bytes_per_second": 5279063.544478442,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.45229118682138,
+    "estimated_full_shards_per_hour": 78.51113369463245,
     "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": 1.2650147490003292,
-      "bytes": 6116564,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0610393269998895,
       "bytes": 5051210,
@@ -562,6 +556,12 @@
       "bytes": 10565027,
       "pages": 23,
       "shards": 3
+    },
+    {
+      "seconds": 1.3083442860006471,
+      "bytes": 8430845,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 618400, "loss": 0.6116386651992798, "lr": 0.0001, "grad_norm": 0.1886418021685392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90516274835, "tokens_per_second": 151838.4977456468, "data_wait_s": 1.375700048811268e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618500, "loss": 0.5130348801612854, "lr": 0.0001, "grad_norm": 0.13897861938266662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90530910109, "tokens_per_second": 118051.8214918686, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 618600, "loss": 0.5336444973945618, "lr": 0.0001, "grad_norm": 0.1612524868048308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90545772982, "tokens_per_second": 154279.9954747513, "data_wait_s": 1.6031000996008515e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 618700, "loss": 0.5049571990966797, "lr": 0.0001, "grad_norm": 0.16050327764789268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90560455586, "tokens_per_second": 153288.01241546977, "data_wait_s": 1.1805999747593887e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6adee8d80f5d30abf10f6d04f1d58733a2bf819ffba92d14b12a0711fc25ed56
-size 303270
+oid sha256:c4456a5ff3c6df99b9965f0a048d73f60fc0d24eaacbe0970a207376651488ce
+size 305388