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": 75440.0863967254,
-      "required_pages_per_second_upper_estimate": 0.006538955487717231
+      "required_token_positions_per_second": 75701.76898196172,
+      "required_pages_per_second_upper_estimate": 0.006561637470977067
     },
     {
       "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": 9430.010799590675,
-      "required_pages_per_second_upper_estimate": 0.0037617722991824935
+      "required_token_positions_per_second": 9462.721122745215,
+      "required_pages_per_second_upper_estimate": 0.0037748209361517533
     },
     {
       "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": 18860.02159918135,
-      "required_pages_per_second_upper_estimate": 0.004362934071927107
+      "required_token_positions_per_second": 18925.44224549043,
+      "required_pages_per_second_upper_estimate": 0.004378067987086663
     },
     {
       "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": 28290.032398772026,
-      "required_pages_per_second_upper_estimate": 0.001722714546756211
+      "required_token_positions_per_second": 28388.163368235648,
+      "required_pages_per_second_upper_estimate": 0.001728690207943139
     },
     {
       "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": 28290.032398772026,
-      "required_pages_per_second_upper_estimate": 0.00217784500587933
+      "required_token_positions_per_second": 28388.163368235648,
+      "required_pages_per_second_upper_estimate": 0.0021853993995525492
     },
     {
       "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": 28290.032398772026,
-      "required_pages_per_second_upper_estimate": 0.3512675217449374
+      "required_token_positions_per_second": 28388.163368235648,
+      "required_pages_per_second_upper_estimate": 0.3524859799624477
     }
   ],
   "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": 0.09977969827343713,
-    "producer_p95_seconds": 0.3288669440007652,
-    "gpu_update_seconds": 0.9773053494700171,
-    "producer_batches_per_second": 10.022078812661785,
-    "gpu_batches_per_second": 1.0232216579416964,
-    "producer_headroom": 9.79463123642448,
+    "producer_batch_seconds": 0.38964838026578263,
+    "producer_p95_seconds": 0.33846487600021646,
+    "gpu_update_seconds": 0.9739270428088406,
+    "producer_batches_per_second": 2.5664164170729804,
+    "gpu_batches_per_second": 1.026770955158986,
+    "producer_headroom": 2.4995023516959476,
     "sustainable": true,
-    "buffer_seconds": 60.59293166714106,
+    "buffer_seconds": 60.38347665414812,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418398.62041856704,
-    "upload_bytes_per_second": 3431159.4409802807,
+    "generation_bytes_per_second": 380931.01699262677,
+    "upload_bytes_per_second": 3510608.2477676445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.77860412042388,
+    "estimated_full_shards_per_hour": 81.73892862638571,
     "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": 0.7924021649996575,
-      "bytes": 1122256,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.1671633029982331,
       "bytes": 11363880,
@@ -562,6 +556,12 @@
       "bytes": 1064415,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.0851846760015178,
+      "bytes": 11839223,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -266,3 +266,4 @@
 {"step": 66600, "loss": 0.5865750210359693, "lr": 0.0003, "grad_norm": 0.1314899339702683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9747201222, "tokens_per_second": 150907.16278123777, "data_wait_s": 1.5180001355474815e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66700, "loss": 0.6159664392471313, "lr": 0.0003, "grad_norm": 0.1255530404184478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9761906396, "tokens_per_second": 149765.30137521963, "data_wait_s": 7.580001692986116e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 66800, "loss": 0.6008422300219536, "lr": 0.0003, "grad_norm": 0.1434360788227507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9776487304, "tokens_per_second": 148791.11147907007, "data_wait_s": 1.032299769576639e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 66900, "loss": 0.608786091208458, "lr": 0.0003, "grad_norm": 0.14827182531631294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9791009949, "tokens_per_second": 148319.05762522173, "data_wait_s": 8.881001122063026e-06, "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 sha256:2011d9738fb7187404dca94ad3b56a052fa9028f8d0450a740c333f9491da4fc
-size 571704
+oid sha256:37e75cd5e652bf45697a643faebbef51fbdff0f02b481a206f19e0d57ea733bf
+size 573822