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": 75941.8898649761,
-      "required_pages_per_second_upper_estimate": 0.005253836150320459
+      "required_token_positions_per_second": 76527.83624133744,
+      "required_pages_per_second_upper_estimate": 0.005294373280220044
     },
     {
       "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": 9492.736233122012,
-      "required_pages_per_second_upper_estimate": 0.003974410429934055
+      "required_token_positions_per_second": 9565.97953016718,
+      "required_pages_per_second_upper_estimate": 0.004005075868912899
     },
     {
       "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": 18985.472466244024,
-      "required_pages_per_second_upper_estimate": 0.003954173234140385
+      "required_token_positions_per_second": 19131.95906033436,
+      "required_pages_per_second_upper_estimate": 0.0039846825285781325
     },
     {
       "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": 28478.208699366034,
-      "required_pages_per_second_upper_estimate": 0.0017072251278200982
+      "required_token_positions_per_second": 28697.93859050154,
+      "required_pages_per_second_upper_estimate": 0.0017203975992855546
     },
     {
       "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": 28478.208699366034,
-      "required_pages_per_second_upper_estimate": 0.0022140554639737795
+      "required_token_positions_per_second": 28697.93859050154,
+      "required_pages_per_second_upper_estimate": 0.0022311384965199156
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28478.208699366034,
-      "required_pages_per_second_upper_estimate": 0.36528788367730064
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28697.93859050154,
+      "required_pages_per_second_upper_estimate": 0.36810634279321125
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10245256441433526,
-    "producer_p95_seconds": 0.312048264000623,
-    "gpu_update_seconds": 0.970847580052164,
-    "producer_batches_per_second": 9.76061463874963,
-    "gpu_batches_per_second": 1.030027803073135,
-    "producer_headroom": 9.476069101851804,
+    "producer_batch_seconds": 0.7987209683441279,
+    "producer_p95_seconds": 0.32224708200010355,
+    "gpu_update_seconds": 0.9634141460303686,
+    "producer_batches_per_second": 1.2520016872389799,
+    "gpu_batches_per_second": 1.0379752094365429,
+    "producer_headroom": 1.2061961363399225,
     "sustainable": true,
-    "buffer_seconds": 60.192549963234164,
+    "buffer_seconds": 60.69509119991322,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320101.2234543803,
-    "upload_bytes_per_second": 5330172.300037838,
+    "generation_bytes_per_second": 263140.13571134186,
+    "upload_bytes_per_second": 5378998.724636954,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.68627097819859,
+    "estimated_full_shards_per_hour": 56.463747534801406,
     "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": 1.2372247309976956,
-      "bytes": 2619840,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 2.7946544099977473,
       "bytes": 9699006,
@@ -562,6 +556,12 @@
       "bytes": 2763568,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.0716314189994591,
+      "bytes": 5628326,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -309,3 +309,4 @@
 {"step": 471100, "loss": 0.55875563621521, "lr": 0.0003, "grad_norm": 0.14749087728292748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68954730461, "tokens_per_second": 151707.98058538302, "data_wait_s": 1.2439999409252778e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 471200, "loss": 0.6057509779930115, "lr": 0.0003, "grad_norm": 0.1524293216732847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68969372295, "tokens_per_second": 152819.6981442068, "data_wait_s": 1.2841999705415219e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 471300, "loss": 0.5933566689491272, "lr": 0.0003, "grad_norm": 0.15876005156108036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68983851162, "tokens_per_second": 150041.98257796268, "data_wait_s": 1.570899985381402e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 471400, "loss": 0.5182668566703796, "lr": 0.0003, "grad_norm": 0.15452290952364842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68998336149, "tokens_per_second": 96471.94536990493, "data_wait_s": 1.16259980131872e-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 sha256:85f711692d1e17f679ad833381c2a78247357e8f8d0c3c8f0548d14b9ff4d3e8
-size 663561
+oid sha256:1fdd3969da20cbc3c2f788c513107c7c2d1c148c836d5e96ed669a187b6696b1
+size 665679