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": 76305.44438263024,
-      "required_pages_per_second_upper_estimate": 0.005278987695414474
+      "required_token_positions_per_second": 76423.47212490725,
+      "required_pages_per_second_upper_estimate": 0.00528715312848729
     },
     {
       "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": 9538.18054782878,
-      "required_pages_per_second_upper_estimate": 0.003993437015516575
+      "required_token_positions_per_second": 9552.934015613406,
+      "required_pages_per_second_upper_estimate": 0.003999613984390556
     },
     {
       "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": 19076.36109565756,
-      "required_pages_per_second_upper_estimate": 0.003973102938752883
+      "required_token_positions_per_second": 19105.86803122681,
+      "required_pages_per_second_upper_estimate": 0.003979248455281738
     },
     {
       "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": 28614.541643486344,
-      "required_pages_per_second_upper_estimate": 0.0017153980796517584
+      "required_token_positions_per_second": 28658.802046840214,
+      "required_pages_per_second_upper_estimate": 0.0017180514232511001
     },
     {
       "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": 28614.541643486344,
-      "required_pages_per_second_upper_estimate": 0.002224654750713887
+      "required_token_positions_per_second": 28658.802046840214,
+      "required_pages_per_second_upper_estimate": 0.0022280958023937103
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28614.541643486344,
-      "required_pages_per_second_upper_estimate": 0.3670366163015654
+      "required_token_positions_per_second": 28658.802046840214,
+      "required_pages_per_second_upper_estimate": 0.3676043412326704
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12829070506245444,
-    "producer_p95_seconds": 0.33443856399389915,
-    "gpu_update_seconds": 0.9662220120270086,
-    "producer_batches_per_second": 7.794796977015446,
-    "gpu_batches_per_second": 1.0349588268043382,
-    "producer_headroom": 7.531504418473909,
+    "producer_batch_seconds": 0.10785460461704588,
+    "producer_p95_seconds": 0.3238576000003377,
+    "gpu_update_seconds": 0.9647297871980779,
+    "producer_batches_per_second": 9.271741373960356,
+    "gpu_batches_per_second": 1.0365596805136073,
+    "producer_headroom": 8.944725082656388,
     "sustainable": true,
-    "buffer_seconds": 60.87198675770154,
+    "buffer_seconds": 60.777976593478904,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314310.9067511513,
-    "upload_bytes_per_second": 4305794.091178219,
+    "generation_bytes_per_second": 314513.7996203542,
+    "upload_bytes_per_second": 4252379.694270034,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.44380380535988,
+    "estimated_full_shards_per_hour": 67.48733989198655,
     "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": 1.7263165549957193,
-      "bytes": 7465725,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 2.012805174999812,
       "bytes": 4595806,
@@ -562,6 +556,12 @@
       "bytes": 7298137,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.4694450449969736,
+      "bytes": 1110429,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -411,3 +411,4 @@
 {"step": 841100, "loss": 0.5524387359619141, "lr": 3e-05, "grad_norm": 0.30640263690148817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123112889526, "tokens_per_second": 151475.88977120374, "data_wait_s": 8.356000762432814e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 841200, "loss": 0.5484036207199097, "lr": 3e-05, "grad_norm": 0.29467330861504815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123127511105, "tokens_per_second": 151440.8779965469, "data_wait_s": 8.506998710799962e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 841300, "loss": 0.525081992149353, "lr": 3e-05, "grad_norm": 0.3052078514605411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123142184039, "tokens_per_second": 152358.17913826404, "data_wait_s": 8.457995136268437e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 841400, "loss": 0.5326145887374878, "lr": 3e-05, "grad_norm": 0.30268375799144104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123156957077, "tokens_per_second": 153413.4737340514, "data_wait_s": 7.086004188749939e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b6b3d93565e8bc62c4e7ba1e2cfc482243fce439fe5a1c0b7f66a843570c3d3
-size 881163
+oid sha256:62437be4fa2c33ae3873e5c398ed1d5b5da4063aa28fc26cd50c5cb2901d54e9
+size 883281