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": 74432.03401940696,
-      "required_pages_per_second_upper_estimate": 0.006451580062536661
+      "required_token_positions_per_second": 74774.21904885587,
+      "required_pages_per_second_upper_estimate": 0.0064812397936292645
     },
     {
       "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": 9304.00425242587,
-      "required_pages_per_second_upper_estimate": 0.0037115064035526846
+      "required_token_positions_per_second": 9346.777381106984,
+      "required_pages_per_second_upper_estimate": 0.0037285692440988446
     },
     {
       "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": 18608.00850485174,
-      "required_pages_per_second_upper_estimate": 0.004304635277832929
+      "required_token_positions_per_second": 18693.554762213967,
+      "required_pages_per_second_upper_estimate": 0.004324424898910972
     },
     {
       "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": 27912.01275727761,
-      "required_pages_per_second_upper_estimate": 0.0016996951339049815
+      "required_token_positions_per_second": 28040.332143320953,
+      "required_pages_per_second_upper_estimate": 0.0017075091112752332
     },
     {
       "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": 27912.01275727761,
-      "required_pages_per_second_upper_estimate": 0.002148744007451748
+      "required_token_positions_per_second": 28040.332143320953,
+      "required_pages_per_second_upper_estimate": 0.0021586223890001596
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27912.01275727761,
-      "required_pages_per_second_upper_estimate": 0.3465737829479321
+      "required_token_positions_per_second": 28040.332143320953,
+      "required_pages_per_second_upper_estimate": 0.3481670802652315
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.2663704616948621,
-    "producer_p95_seconds": 0.40725806199770886,
-    "gpu_update_seconds": 0.9905412497631948,
-    "producer_batches_per_second": 3.7541700143371735,
-    "gpu_batches_per_second": 1.0095490725288487,
-    "producer_headroom": 3.7186602578250545,
+    "producer_batch_seconds": 0.12483373178133661,
+    "producer_p95_seconds": 0.3749859599993215,
+    "gpu_update_seconds": 0.9860082918663143,
+    "producer_batches_per_second": 8.010655339148533,
+    "gpu_batches_per_second": 1.0141902540263654,
+    "producer_headroom": 7.898572587683616,
     "sustainable": true,
-    "buffer_seconds": 60.42301623555488,
+    "buffer_seconds": 60.14650580384517,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450750.01889814826,
-    "upload_bytes_per_second": 5787306.752317857,
+    "generation_bytes_per_second": 478362.9079638481,
+    "upload_bytes_per_second": 5863565.803500393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.72046113212906,
+    "estimated_full_shards_per_hour": 102.64554433046897,
     "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": 0.8043565109983319,
-      "bytes": 540912,
-      "pages": 18,
-      "shards": 1
-    },
     {
       "seconds": 1.4004500939990976,
       "bytes": 16241287,
@@ -562,6 +556,12 @@
       "bytes": 5767450,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1749305560006178,
+      "bytes": 8197083,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 395500, "loss": 0.6386530287563801, "lr": 0.0003, "grad_norm": 0.15998983031841527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57889856109, "tokens_per_second": 148535.13742564007, "data_wait_s": 9.37400182010606e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 395600, "loss": 0.5625989250838757, "lr": 0.0003, "grad_norm": 0.17024057662099099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57904399230, "tokens_per_second": 147798.79953583484, "data_wait_s": 1.2132004485465586e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395700, "loss": 0.6006899625062943, "lr": 0.0003, "grad_norm": 0.16791024958231635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57919014382, "tokens_per_second": 148759.88234629392, "data_wait_s": 1.3379001757130027e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 395800, "loss": 0.6065213084220886, "lr": 0.0003, "grad_norm": 0.214611582458161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57933679130, "tokens_per_second": 150357.68617321388, "data_wait_s": 1.2053998943883926e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51c2804d02a4babef845d420a8f4cc4cf382ff5ccab365c8343b097c8a758716
-size 868455
+oid sha256:858dc64ddadd101452e45cc152cd852b57a316f3cbbb6f314bcb972f6ec667d7
+size 870573