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": 75235.71873057302,
-      "required_pages_per_second_upper_estimate": 0.006521241416380257
+      "required_token_positions_per_second": 75270.4590666937,
+      "required_pages_per_second_upper_estimate": 0.006524252620666621
     },
     {
       "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": 9404.464841321627,
-      "required_pages_per_second_upper_estimate": 0.0037515816344828577
+      "required_token_positions_per_second": 9408.807383336713,
+      "required_pages_per_second_upper_estimate": 0.003753313939419464
     },
     {
       "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": 18808.929682643255,
-      "required_pages_per_second_upper_estimate": 0.004351114856223039
+      "required_token_positions_per_second": 18817.614766673425,
+      "required_pages_per_second_upper_estimate": 0.004353123997561161
     },
     {
       "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": 28213.39452396488,
-      "required_pages_per_second_upper_estimate": 0.00171804770226831
+      "required_token_positions_per_second": 28226.42215001014,
+      "required_pages_per_second_upper_estimate": 0.001718841016343795
     },
     {
       "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": 28213.39452396488,
-      "required_pages_per_second_upper_estimate": 0.00217194521012946
+      "required_token_positions_per_second": 28226.42215001014,
+      "required_pages_per_second_upper_estimate": 0.0021729481128452016
     },
     {
       "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": 28213.39452396488,
-      "required_pages_per_second_upper_estimate": 0.3503159358303001
+      "required_token_positions_per_second": 28226.42215001014,
+      "required_pages_per_second_upper_estimate": 0.35047769534512263
     }
   ],
   "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": 0.1412360786407305,
-    "producer_p95_seconds": 0.25788864299829584,
-    "gpu_update_seconds": 0.979960067425257,
-    "producer_batches_per_second": 7.080343844321475,
-    "gpu_batches_per_second": 1.0204497440670168,
-    "producer_headroom": 6.9384542310752755,
+    "producer_batch_seconds": 0.16477561939845486,
+    "producer_p95_seconds": 0.3243474350019824,
+    "gpu_update_seconds": 0.9795077765458159,
+    "producer_batches_per_second": 6.0688589953458685,
+    "gpu_batches_per_second": 1.0209209400321955,
+    "producer_headroom": 5.944494580701305,
     "sustainable": true,
-    "buffer_seconds": 60.757524180365934,
+    "buffer_seconds": 60.729482145840585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538993.2233139698,
-    "upload_bytes_per_second": 5078230.556730361,
+    "generation_bytes_per_second": 538552.9021787721,
+    "upload_bytes_per_second": 5156047.632222512,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.65539860309906,
+    "estimated_full_shards_per_hour": 115.56091593763708,
     "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": 0.8234439619991463,
-      "bytes": 1074472,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.0881272910010011,
       "bytes": 13091178,
@@ -562,6 +556,12 @@
       "bytes": 13578331,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.2704945900004532,
+      "bytes": 9798343,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -223,3 +223,4 @@
 {"step": 62300, "loss": 0.6850302442908287, "lr": 0.0003, "grad_norm": 0.145835814374586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9118026102, "tokens_per_second": 149893.7493081312, "data_wait_s": 8.742001227801666e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 62400, "loss": 0.6482821516692638, "lr": 0.0003, "grad_norm": 0.13581931887468313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9132514092, "tokens_per_second": 147636.5535333202, "data_wait_s": 1.8000999261857942e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 62500, "loss": 0.6781617663800716, "lr": 0.0003, "grad_norm": 0.14668453708861534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9146982355, "tokens_per_second": 147447.39687179116, "data_wait_s": 9.16399949346669e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 62600, "loss": 0.5980433002114296, "lr": 0.0003, "grad_norm": 0.13367296387781344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9161457722, "tokens_per_second": 147594.46401359976, "data_wait_s": 1.0152001777896658e-05, "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 sha256:294fcd6a3ce55c87af7bb588cbc1cf444f4d17dbf6383af07138751cf78cd91b
-size 479847
+oid sha256:1bc99ef9f7f0d75f531e51ce5aa226d9aadcf8a0cc08025cde94f920c1784203
+size 481965