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": 74270.01868695545,
-      "required_pages_per_second_upper_estimate": 0.0064375369841438695
+      "required_token_positions_per_second": 73782.08127499156,
+      "required_pages_per_second_upper_estimate": 0.006395243805940905
     },
     {
       "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": 9283.75233586943,
-      "required_pages_per_second_upper_estimate": 0.0037034276112451856
+      "required_token_positions_per_second": 9222.760159373945,
+      "required_pages_per_second_upper_estimate": 0.0036790969201268334
     },
     {
       "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": 18567.50467173886,
-      "required_pages_per_second_upper_estimate": 0.004295265428885381
+      "required_token_positions_per_second": 18445.52031874789,
+      "required_pages_per_second_upper_estimate": 0.004267046495672208
     },
     {
       "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": 27851.257007608296,
-      "required_pages_per_second_upper_estimate": 0.0016959954275109035
+      "required_token_positions_per_second": 27668.280478121833,
+      "required_pages_per_second_upper_estimate": 0.0016848531168688352
     },
     {
       "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": 27851.257007608296,
-      "required_pages_per_second_upper_estimate": 0.002144066861659525
+      "required_token_positions_per_second": 27668.280478121833,
+      "required_pages_per_second_upper_estimate": 0.002129980821908205
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27851.257007608296,
-      "required_pages_per_second_upper_estimate": 0.3458193998734531
+      "required_token_positions_per_second": 27668.280478121833,
+      "required_pages_per_second_upper_estimate": 0.343547443760282
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12125674908628525,
-    "producer_p95_seconds": 0.40255499700288055,
-    "gpu_update_seconds": 0.9927020526379556,
-    "producer_batches_per_second": 8.246963633244107,
-    "gpu_batches_per_second": 1.0073515989441657,
-    "producer_headroom": 8.186777726751997,
+    "producer_batch_seconds": 0.09595149529707214,
+    "producer_p95_seconds": 0.37530223399517126,
+    "gpu_update_seconds": 0.9992670134257938,
+    "producer_batches_per_second": 10.421932424335173,
+    "gpu_batches_per_second": 1.0007335242376243,
+    "producer_headroom": 10.414293287790851,
     "sustainable": true,
-    "buffer_seconds": 60.5548252109153,
+    "buffer_seconds": 60.955287818973424,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 471010.4773408362,
-    "upload_bytes_per_second": 3809018.795322353,
+    "generation_bytes_per_second": 461707.7646010129,
+    "upload_bytes_per_second": 3835803.6111592688,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.06788387459578,
+    "estimated_full_shards_per_hour": 99.07173827669898,
     "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": 0.8995478449978691,
-      "bytes": 2436827,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.0387542439966637,
       "bytes": 11801776,
@@ -562,6 +556,12 @@
       "bytes": 13452885,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.0817572929954622,
+      "bytes": 6130820,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 37500, "loss": 0.6066446155309677, "lr": 0.0003, "grad_norm": 0.1288046809234849, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5487480301, "tokens_per_second": 149562.4722017392, "data_wait_s": 9.538998710922897e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 37600, "loss": 0.6831776015460491, "lr": 0.0003, "grad_norm": 0.13815717938820057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5502116465, "tokens_per_second": 148578.9497098662, "data_wait_s": 1.345400232821703e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37700, "loss": 0.6304601151496172, "lr": 0.0003, "grad_norm": 0.1387206411435085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5516753921, "tokens_per_second": 148644.81277852427, "data_wait_s": 1.16259980131872e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 37800, "loss": 0.7046154029667377, "lr": 0.0003, "grad_norm": 0.155621582148506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5531510173, "tokens_per_second": 149407.14831093606, "data_wait_s": 1.4583994925487787e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dce732ae14bd7c5a97c0b4c79bf6ef6fbcb886c2278f129859621fcc3a2f6ee
-size 798358
+oid sha256:d103ba282550a5c050a03757220df79a31158e3facd36d44fb0555d45064e345
+size 800476