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": 74731.14307033684,
-      "required_pages_per_second_upper_estimate": 0.0064775060769863045
+      "required_token_positions_per_second": 75388.66727303428,
+      "required_pages_per_second_upper_estimate": 0.006534498608396263
     },
     {
       "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": 9341.392883792105,
-      "required_pages_per_second_upper_estimate": 0.0037264212876145304
+      "required_token_positions_per_second": 9423.583409129285,
+      "required_pages_per_second_upper_estimate": 0.003759208317029394
     },
     {
       "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": 18682.78576758421,
-      "required_pages_per_second_upper_estimate": 0.00432193368153121
+      "required_token_positions_per_second": 18847.16681825857,
+      "required_pages_per_second_upper_estimate": 0.00435996034459704
     },
     {
       "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": 28024.17865137631,
-      "required_pages_per_second_upper_estimate": 0.0017065254483666268
+      "required_token_positions_per_second": 28270.750227387856,
+      "required_pages_per_second_upper_estimate": 0.00172154036368465
     },
     {
       "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": 28024.17865137631,
-      "required_pages_per_second_upper_estimate": 0.0021573788484745334
+      "required_token_positions_per_second": 28270.750227387856,
+      "required_pages_per_second_upper_estimate": 0.00217636061094971
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28024.17865137631,
-      "required_pages_per_second_upper_estimate": 0.3479665079575389
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28270.750227387856,
+      "required_pages_per_second_upper_estimate": 0.35102810170962234
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09391472234378284,
-    "producer_p95_seconds": 0.29560095299893874,
-    "gpu_update_seconds": 0.9865766395491545,
-    "producer_batches_per_second": 10.64795779664252,
-    "gpu_batches_per_second": 1.0136059986753585,
-    "producer_headroom": 10.505026421072797,
+    "producer_batch_seconds": 0.11882688294539889,
+    "producer_p95_seconds": 0.31474311299825786,
+    "gpu_update_seconds": 0.9779719242546115,
+    "producer_batches_per_second": 8.415604072182061,
+    "gpu_batches_per_second": 1.0225242414419797,
+    "producer_headroom": 8.230224508236835,
     "sustainable": true,
-    "buffer_seconds": 60.18117501249843,
+    "buffer_seconds": 60.63425930378592,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432655.63369797415,
-    "upload_bytes_per_second": 4101894.7906612535,
+    "generation_bytes_per_second": 552428.0613550623,
+    "upload_bytes_per_second": 4056353.9797395477,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.83782728390139,
+    "estimated_full_shards_per_hour": 118.53820209969427,
     "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": 1.247284810997371,
-      "bytes": 6017710,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9974236470006872,
       "bytes": 6556698,
@@ -562,6 +556,12 @@
       "bytes": 6024195,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.20260026500182,
+      "bytes": 1126978,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 103900, "loss": 0.6776418164372444, "lr": 0.0003, "grad_norm": 0.1505266028600515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15204204976, "tokens_per_second": 148174.34915305296, "data_wait_s": 1.818699820432812e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104000, "loss": 0.5718474229797721, "lr": 0.0003, "grad_norm": 0.13511212981686685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15218788855, "tokens_per_second": 147903.42706381212, "data_wait_s": 1.5044999599922448e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104100, "loss": 0.5915816277265549, "lr": 0.0003, "grad_norm": 0.1432474375497813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15233434216, "tokens_per_second": 148176.27755571835, "data_wait_s": 1.3013999705435708e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 104200, "loss": 0.5905342511832714, "lr": 0.0003, "grad_norm": 0.13090020853107528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15248055277, "tokens_per_second": 147394.1055627227, "data_wait_s": 1.8148999515688047e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6087d4df06da5a15a3ac203c1f47ed49872a2701b6ea2ce5fe008ca1f56a95ba
-size 407052
+oid sha256:893ea86c45da6a23f983f3a1263dc972126cecaaa9545d297fb52252147b2555
+size 409170