Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76081.1847816869,
-      "required_pages_per_second_upper_estimate": 0.005263472895867248
+      "required_token_positions_per_second": 76707.95162330984,
+      "required_pages_per_second_upper_estimate": 0.00530683408027016
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9510.148097710862,
-      "required_pages_per_second_upper_estimate": 0.003981700414036327
+      "required_token_positions_per_second": 9588.49395291373,
+      "required_pages_per_second_upper_estimate": 0.00401450218756227
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19020.296195421724,
-      "required_pages_per_second_upper_estimate": 0.003961426098564604
+      "required_token_positions_per_second": 19176.98790582746,
+      "required_pages_per_second_upper_estimate": 0.003994060849603834
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28530.444293132587,
-      "required_pages_per_second_upper_estimate": 0.0017103565719072695
+      "required_token_positions_per_second": 28765.48185874119,
+      "required_pages_per_second_upper_estimate": 0.0017244467150839226
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28530.444293132587,
-      "required_pages_per_second_upper_estimate": 0.0022181165516290337
+      "required_token_positions_per_second": 28765.48185874119,
+      "required_pages_per_second_upper_estimate": 0.002236389688534109
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28530.444293132587,
-      "required_pages_per_second_upper_estimate": 0.36595790578792714
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28765.48185874119,
+      "required_pages_per_second_upper_estimate": 0.36897271531587833
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4321026642421941,
-    "producer_p95_seconds": 0.2735124450009607,
-    "gpu_update_seconds": 0.9690700823279855,
-    "producer_batches_per_second": 2.314264832765527,
-    "gpu_batches_per_second": 1.0319171112967516,
-    "producer_headroom": 2.2426848120168508,
+    "producer_batch_seconds": 0.09184869667186035,
+    "producer_p95_seconds": 0.2711483999992197,
+    "gpu_update_seconds": 0.9611519854167987,
+    "producer_batches_per_second": 10.88747076697899,
+    "gpu_batches_per_second": 1.0404181806547017,
+    "producer_headroom": 10.464514143849211,
     "sustainable": true,
-    "buffer_seconds": 60.082345104335104,
+    "buffer_seconds": 60.55257508125832,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344755.98417331884,
-    "upload_bytes_per_second": 5207063.29804664,
+    "generation_bytes_per_second": 344124.52596430195,
+    "upload_bytes_per_second": 5306241.539833224,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.9766086950271,
+    "estimated_full_shards_per_hour": 73.84111246296686,
     "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": 1.1655933139991248,
-      "bytes": 2619678,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0188674239998363,
       "bytes": 6227206,
@@ -562,6 +556,12 @@
       "bytes": 7624893,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1967464800000016,
+      "bytes": 10837616,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 627300, "loss": 0.5515968799591064, "lr": 0.0001, "grad_norm": 0.1660719319697747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91819960425, "tokens_per_second": 151817.48481567745, "data_wait_s": 1.1936001101275906e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627400, "loss": 0.5578224062919617, "lr": 0.0001, "grad_norm": 0.18071927340125216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91834666325, "tokens_per_second": 151007.69225771772, "data_wait_s": 1.364099807688035e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 627500, "loss": 0.5221641063690186, "lr": 0.0001, "grad_norm": 0.15815645773418807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91849452402, "tokens_per_second": 152911.36024273763, "data_wait_s": 1.022600190481171e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 627600, "loss": 0.5134794116020203, "lr": 0.0001, "grad_norm": 0.1619216194800924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91864129114, "tokens_per_second": 152124.95722376273, "data_wait_s": 1.1240997991990298e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adf03d9c21836c8b96ff1d8ca7d7e99ee88f4e5b6451e8ef52a24555eccf600f
-size 493338
+oid sha256:356e9e4cb6d4ae023d3e5b3dc667df7b01b0c8604ffff583471e6bf427759174
+size 495456