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": 74814.00240683355,
-      "required_pages_per_second_upper_estimate": 0.00648468811426877
+      "required_token_positions_per_second": 75074.2737388551,
+      "required_pages_per_second_upper_estimate": 0.006507247773676739
     },
     {
       "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": 9351.750300854193,
-      "required_pages_per_second_upper_estimate": 0.0037305530161377827
+      "required_token_positions_per_second": 9384.284217356888,
+      "required_pages_per_second_upper_estimate": 0.0037435312818561066
     },
     {
       "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": 18703.500601708387,
-      "required_pages_per_second_upper_estimate": 0.004326725693837209
+      "required_token_positions_per_second": 18768.568434713776,
+      "required_pages_per_second_upper_estimate": 0.004341777991848257
     },
     {
       "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": 28055.250902562584,
-      "required_pages_per_second_upper_estimate": 0.001708417585440368
+      "required_token_positions_per_second": 28152.852652070662,
+      "required_pages_per_second_upper_estimate": 0.0017143610199086055
     },
     {
       "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": 28055.250902562584,
-      "required_pages_per_second_upper_estimate": 0.0021597708763843493
+      "required_token_positions_per_second": 28152.852652070662,
+      "required_pages_per_second_upper_estimate": 0.0021672845292404156
     },
     {
       "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": 28055.250902562584,
-      "required_pages_per_second_upper_estimate": 0.3483523213251373
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28152.852652070662,
+      "required_pages_per_second_upper_estimate": 0.34956420840198493
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12824511989052212,
-    "producer_p95_seconds": 0.3036303160006355,
-    "gpu_update_seconds": 0.9854839686168915,
-    "producer_batches_per_second": 7.7975676645915355,
-    "gpu_batches_per_second": 1.0147298503531026,
-    "producer_headroom": 7.684377927660412,
+    "producer_batch_seconds": 0.0746424604766105,
+    "producer_p95_seconds": 0.2696097420011938,
+    "gpu_update_seconds": 0.9820674423899445,
+    "producer_batches_per_second": 13.397200381857106,
+    "gpu_batches_per_second": 1.0182600062236207,
+    "producer_headroom": 13.156954314195994,
     "sustainable": true,
-    "buffer_seconds": 60.11452208563038,
+    "buffer_seconds": 60.88818142817656,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288156.9093198548,
-    "upload_bytes_per_second": 5134820.860751925,
+    "generation_bytes_per_second": 288136.2639128291,
+    "upload_bytes_per_second": 5097572.210081281,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.83176479050382,
+    "estimated_full_shards_per_hour": 61.82733476675658,
     "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": 1.2958617730000697,
-      "bytes": 7155083,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0062387039997702,
       "bytes": 10437075,
@@ -562,6 +556,12 @@
       "bytes": 6193785,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9691767799995432,
+      "bytes": 2430413,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 259700, "loss": 0.6227231193333864, "lr": 0.0003, "grad_norm": 0.14113326448330837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38012973655, "tokens_per_second": 97217.35659860563, "data_wait_s": 1.955900006578304e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 259800, "loss": 0.6153518669307232, "lr": 0.0003, "grad_norm": 0.15683379856812443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38027554787, "tokens_per_second": 147895.20709408604, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 259900, "loss": 0.6418608222156763, "lr": 0.0003, "grad_norm": 0.1562708766746261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38042090249, "tokens_per_second": 148173.24023632068, "data_wait_s": 1.5534002159256488e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 260000, "loss": 0.586906854994595, "lr": 0.0003, "grad_norm": 0.14958846110889634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38056709872, "tokens_per_second": 149146.71735342673, "data_wait_s": 1.6132002201629803e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2bcc930d2297a717e144fde8c5b8cbdecf1ce629704663aaa01a2bb670d4f8b3
-size 486201
+oid sha256:fe143b3512842c8a0dcdf0810b0b5df0fdb2b6c13e5c4bdeac4bb96929dd0918
+size 488319