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": 76385.16045051531,
-      "required_pages_per_second_upper_estimate": 0.005284502637957524
+      "required_token_positions_per_second": 76245.62738112485,
+      "required_pages_per_second_upper_estimate": 0.0052748494164557656
     },
     {
       "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": 9548.145056314414,
-      "required_pages_per_second_upper_estimate": 0.00399760894713691
+      "required_token_positions_per_second": 9530.703422640607,
+      "required_pages_per_second_upper_estimate": 0.003990306499340416
     },
     {
       "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": 19096.290112628827,
-      "required_pages_per_second_upper_estimate": 0.003977253627424477
+      "required_token_positions_per_second": 19061.406845281213,
+      "required_pages_per_second_upper_estimate": 0.003969988362769592
     },
     {
       "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": 28644.43516894324,
-      "required_pages_per_second_upper_estimate": 0.0017171901508581289
+      "required_token_positions_per_second": 28592.110267921824,
+      "required_pages_per_second_upper_estimate": 0.0017140533529373927
     },
     {
       "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": 28644.43516894324,
-      "required_pages_per_second_upper_estimate": 0.002226978840830438
+      "required_token_positions_per_second": 28592.110267921824,
+      "required_pages_per_second_upper_estimate": 0.0022229108099289393
     },
     {
       "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": 28644.43516894324,
-      "required_pages_per_second_upper_estimate": 0.36742005834895963
+      "required_token_positions_per_second": 28592.110267921824,
+      "required_pages_per_second_upper_estimate": 0.3667488907007584
     }
   ],
   "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": 0.26430587567189434,
-    "producer_p95_seconds": 0.240215845999046,
-    "gpu_update_seconds": 0.9652136562279436,
-    "producer_batches_per_second": 3.783495154838654,
-    "gpu_batches_per_second": 1.036040045173005,
-    "producer_headroom": 3.651881191722527,
+    "producer_batch_seconds": 0.06814504658609621,
+    "producer_p95_seconds": 0.22161367900116602,
+    "gpu_update_seconds": 0.9669800424286611,
+    "producer_batches_per_second": 14.674580913765672,
+    "gpu_batches_per_second": 1.034147506796941,
+    "producer_headroom": 14.190026874615949,
     "sustainable": true,
-    "buffer_seconds": 60.80846034236045,
+    "buffer_seconds": 60.91974267300565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 431838.8593536913,
-    "upload_bytes_per_second": 4887187.340630625,
+    "generation_bytes_per_second": 431010.28649540804,
+    "upload_bytes_per_second": 4888653.58262817,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.66256652315191,
+    "estimated_full_shards_per_hour": 92.4847740759533,
     "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": 1.7082208169995283,
-      "bytes": 12327681,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 2.076565071999539,
       "bytes": 6113882,
@@ -562,6 +556,12 @@
       "bytes": 2530940,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.3693337599997903,
+      "bytes": 10798536,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 732100, "loss": 0.52305668592453, "lr": 5e-05, "grad_norm": 0.2194483695417114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107158494165, "tokens_per_second": 150453.51858312136, "data_wait_s": 2.0274999769753776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732200, "loss": 0.5338050723075867, "lr": 5e-05, "grad_norm": 0.22597325297167617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107173208792, "tokens_per_second": 151291.6350583266, "data_wait_s": 2.2498999896924943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732300, "loss": 0.5135024785995483, "lr": 5e-05, "grad_norm": 0.21673142068745865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107187908117, "tokens_per_second": 150848.78265277317, "data_wait_s": 1.97749996004859e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 732400, "loss": 0.5128567814826965, "lr": 5e-05, "grad_norm": 0.21129127978977358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107202673752, "tokens_per_second": 150691.97426144668, "data_wait_s": 2.23299994104309e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f64d2a3df10253277123263ecb3e08579ba54c80219f1139e45562d5fc66d9b1
-size 262245
+oid sha256:4aef2c5453e7294af5fb5c1dd86e822afb4799fb92df567dc3f78f922d89b60a
+size 264363