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": 75123.29236859328,
-      "required_pages_per_second_upper_estimate": 0.0065114965842658626
+      "required_token_positions_per_second": 74555.70822138175,
+      "required_pages_per_second_upper_estimate": 0.0064622998289144366
     },
     {
       "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": 9390.41154607416,
-      "required_pages_per_second_upper_estimate": 0.0037459755648931543
+      "required_token_positions_per_second": 9319.463527672719,
+      "required_pages_per_second_upper_estimate": 0.0037176733395854152
     },
     {
       "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": 18780.82309214832,
-      "required_pages_per_second_upper_estimate": 0.004344612891171134
+      "required_token_positions_per_second": 18638.927055345437,
+      "required_pages_per_second_upper_estimate": 0.004311787740341742
     },
     {
       "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": 28171.23463822248,
-      "required_pages_per_second_upper_estimate": 0.0017154803864224235
+      "required_token_positions_per_second": 27958.390583018154,
+      "required_pages_per_second_upper_estimate": 0.001702519300166934
     },
     {
       "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": 28171.23463822248,
-      "required_pages_per_second_upper_estimate": 0.002168699625418977
+      "required_token_positions_per_second": 27958.390583018154,
+      "required_pages_per_second_upper_estimate": 0.0021523143008592934
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28171.23463822248,
-      "required_pages_per_second_upper_estimate": 0.34979245114944035
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27958.390583018154,
+      "required_pages_per_second_upper_estimate": 0.3471496403270317
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10057266216409744,
-    "producer_p95_seconds": 0.3390224120012135,
-    "gpu_update_seconds": 0.9814266344751338,
-    "producer_batches_per_second": 9.943059858238309,
-    "gpu_batches_per_second": 1.0189248639403385,
-    "producer_headroom": 9.758383773055625,
+    "producer_batch_seconds": 0.08360177253891266,
+    "producer_p95_seconds": 0.33020705899980385,
+    "gpu_update_seconds": 0.9888981240856302,
+    "producer_batches_per_second": 11.961468873576186,
+    "gpu_batches_per_second": 1.0112265112492098,
+    "producer_headroom": 11.828674130388144,
     "sustainable": true,
-    "buffer_seconds": 60.84845133745829,
+    "buffer_seconds": 60.32278556922344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418303.44082475774,
-    "upload_bytes_per_second": 5548369.144216122,
+    "generation_bytes_per_second": 417809.7682826356,
+    "upload_bytes_per_second": 5511460.304690656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.75818079525995,
+    "estimated_full_shards_per_hour": 89.65225015982915,
     "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": 1.238607878000039,
-      "bytes": 6177061,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0433953600004315,
       "bytes": 1237044,
@@ -562,6 +556,12 @@
       "bytes": 7462663,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8068668020023324,
+      "bytes": 982732,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -241,3 +241,4 @@
 {"step": 179100, "loss": 0.5237388759851456, "lr": 0.0003, "grad_norm": 0.15559777964325278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26212625519, "tokens_per_second": 149030.32106352434, "data_wait_s": 1.207899913424626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179200, "loss": 0.6327262502163649, "lr": 0.0003, "grad_norm": 0.15070615208766142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26227355581, "tokens_per_second": 149891.60798435472, "data_wait_s": 1.9053000869462267e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179300, "loss": 0.6335432901978493, "lr": 0.0003, "grad_norm": 0.15791971190942056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26241899685, "tokens_per_second": 147946.1481803719, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 179400, "loss": 0.6733214948326349, "lr": 0.0003, "grad_norm": 0.17627854405043109, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26256400691, "tokens_per_second": 148238.68769730596, "data_wait_s": 1.0067997209262103e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1964e008b9ef276d5391d3000159c331f5204ac8e60b598cca9c86443319087
-size 517971
+oid sha256:d385c8995a462629087db9bb66ac9f1ad25309a2993134e280e4f3841a248f71
+size 520089