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": 76857.74530311573,
-      "required_pages_per_second_upper_estimate": 0.005317197154608354
+      "required_token_positions_per_second": 76845.77707109599,
+      "required_pages_per_second_upper_estimate": 0.005316369164547104
     },
     {
       "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": 9607.218162889467,
-      "required_pages_per_second_upper_estimate": 0.0040223416232731445
+      "required_token_positions_per_second": 9605.722133886999,
+      "required_pages_per_second_upper_estimate": 0.0040217152671704485
     },
     {
       "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": 19214.436325778934,
-      "required_pages_per_second_upper_estimate": 0.00400186036789848
+      "required_token_positions_per_second": 19211.444267773997,
+      "required_pages_per_second_upper_estimate": 0.004001237201121928
     },
     {
       "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": 28821.6544886684,
-      "required_pages_per_second_upper_estimate": 0.001727814178477419
+      "required_token_positions_per_second": 28817.166401660998,
+      "required_pages_per_second_upper_estimate": 0.0017275451245142373
     },
     {
       "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": 28821.6544886684,
-      "required_pages_per_second_upper_estimate": 0.0022407568634336622
+      "required_token_positions_per_second": 28817.166401660998,
+      "required_pages_per_second_upper_estimate": 0.0022404079344098453
     },
     {
       "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": 28821.6544886684,
-      "required_pages_per_second_upper_estimate": 0.36969323749911365
+      "required_token_positions_per_second": 28817.166401660998,
+      "required_pages_per_second_upper_estimate": 0.36963566913791507
     }
   ],
   "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": 0.07224095703887201,
-    "producer_p95_seconds": 0.2611011650005821,
-    "gpu_update_seconds": 0.959278726031157,
-    "producer_batches_per_second": 13.842563013968817,
-    "gpu_batches_per_second": 1.0424498874663048,
-    "producer_headroom": 13.27887621304602,
+    "producer_batch_seconds": 0.12836127027327393,
+    "producer_p95_seconds": 0.24796521400276106,
+    "gpu_update_seconds": 0.9594281274791262,
+    "producer_batches_per_second": 7.7905118722419635,
+    "gpu_batches_per_second": 1.042287557930447,
+    "producer_headroom": 7.474436217689009,
     "sustainable": true,
-    "buffer_seconds": 60.43455973996289,
+    "buffer_seconds": 60.44397203118495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405082.23382658977,
-    "upload_bytes_per_second": 4963828.883437685,
+    "generation_bytes_per_second": 352875.08277493494,
+    "upload_bytes_per_second": 4964521.663588484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.92121754740019,
+    "estimated_full_shards_per_hour": 75.71877825199162,
     "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": 1.1019184790002328,
-      "bytes": 6210962,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.991184726000938,
       "bytes": 5824172,
@@ -562,6 +556,12 @@
       "bytes": 6256940,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1927087240001129,
+      "bytes": 6721845,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -235,3 +235,4 @@
 {"step": 783500, "loss": 0.5408061146736145, "lr": 3e-05, "grad_norm": 0.2619653487019164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114682268048, "tokens_per_second": 149772.36378437467, "data_wait_s": 1.0682000720407814e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783600, "loss": 0.5225484371185303, "lr": 3e-05, "grad_norm": 0.2480367554437786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114696800610, "tokens_per_second": 151193.368052761, "data_wait_s": 1.4572997315553948e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783700, "loss": 0.5481111407279968, "lr": 3e-05, "grad_norm": 0.26958007243315957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114711391951, "tokens_per_second": 140675.79402215322, "data_wait_s": 1.4942001143936068e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 783800, "loss": 0.5824326276779175, "lr": 3e-05, "grad_norm": 0.2746221799751516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114726173008, "tokens_per_second": 153272.93106106776, "data_wait_s": 1.0961997759295627e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a2204cb08a9f7499fbe81abc83d8f3426209bc50f8469d59cb8f37fe7cd121f1
-size 505263
+oid sha256:2a25a7a727eece7b16e261f3fd67a251823a16fb6e8b015bce83726aa607a88f
+size 507381