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": 77322.81940488383,
-      "required_pages_per_second_upper_estimate": 0.0053493720600892615
+      "required_token_positions_per_second": 75261.80351141686,
+      "required_pages_per_second_upper_estimate": 0.0052067861983634825
     },
     {
       "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": 9665.35242561048,
-      "required_pages_per_second_upper_estimate": 0.00404668122509298
+      "required_token_positions_per_second": 9407.725438927107,
+      "required_pages_per_second_upper_estimate": 0.003938818185631899
     },
     {
       "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": 19330.70485122096,
-      "required_pages_per_second_upper_estimate": 0.0040260760355408445
+      "required_token_positions_per_second": 18815.450877854215,
+      "required_pages_per_second_upper_estimate": 0.003918762221049594
     },
     {
       "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": 28996.057276831438,
-      "required_pages_per_second_upper_estimate": 0.0017382693593301548
+      "required_token_positions_per_second": 28223.176316781322,
+      "required_pages_per_second_upper_estimate": 0.0016919363258960452
     },
     {
       "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": 28996.057276831438,
-      "required_pages_per_second_upper_estimate": 0.0022543159130965448
+      "required_token_positions_per_second": 28223.176316781322,
+      "required_pages_per_second_upper_estimate": 0.002194227817996717
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28996.057276831438,
-      "required_pages_per_second_upper_estimate": 0.3719302892065448
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28223.176316781322,
+      "required_pages_per_second_upper_estimate": 0.3620166021059417
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07880365366426645,
-    "producer_p95_seconds": 0.2761755780011299,
-    "gpu_update_seconds": 0.9535089455796955,
-    "producer_batches_per_second": 12.68976695243574,
-    "gpu_batches_per_second": 1.0487578586816926,
-    "producer_headroom": 12.099806306469068,
+    "producer_batch_seconds": 0.2737312463671344,
+    "producer_p95_seconds": 0.29197228199700476,
+    "gpu_update_seconds": 0.9796204257690398,
+    "producer_batches_per_second": 3.653218305442477,
+    "gpu_batches_per_second": 1.0208035415502503,
+    "producer_headroom": 3.57876727180481,
     "sustainable": true,
-    "buffer_seconds": 60.07106357152082,
+    "buffer_seconds": 60.73646639768047,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 559977.8443814401,
-    "upload_bytes_per_second": 4959653.373041274,
+    "generation_bytes_per_second": 518315.41679740796,
+    "upload_bytes_per_second": 4988737.386952134,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.07910489360047,
+    "estimated_full_shards_per_hour": 111.21842267934493,
     "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": 1.272974397001235,
-      "bytes": 8064738,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.1465204630003427,
       "bytes": 10636216,
@@ -562,6 +556,12 @@
       "bytes": 2789154,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.218474695997429,
+      "bytes": 10329696,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 465600, "loss": 0.6022682785987854, "lr": 0.0003, "grad_norm": 0.1557973367140862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68149959809, "tokens_per_second": 153728.81772058288, "data_wait_s": 1.483599771745503e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465700, "loss": 0.592523992061615, "lr": 0.0003, "grad_norm": 0.14624734479052887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68164666775, "tokens_per_second": 152591.0714904238, "data_wait_s": 1.0157000360777602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465800, "loss": 0.5672600269317627, "lr": 0.0003, "grad_norm": 0.15931824950900292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68179393885, "tokens_per_second": 153096.10006626567, "data_wait_s": 1.3978002243675292e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 465900, "loss": 0.6099376678466797, "lr": 0.0003, "grad_norm": 0.15082115595999188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68193861165, "tokens_per_second": 149748.7105596939, "data_wait_s": 1.4932000340195373e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62fc4872d4df064516e97d96ef3b0b42c5db68921ccdf8a6022af40b1e95b69b
-size 546288
+oid sha256:e135fa8e96fbcc47a076eef91a5267f41203ead80f4ec6c10b24d6b61563b5c5
+size 548406