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": 76771.06516405426,
-      "required_pages_per_second_upper_estimate": 0.0053112004214624494
+      "required_token_positions_per_second": 76484.56343355196,
+      "required_pages_per_second_upper_estimate": 0.005291379566970697
     },
     {
       "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": 9596.383145506783,
-      "required_pages_per_second_upper_estimate": 0.004017805227755906
+      "required_token_positions_per_second": 9560.570429193995,
+      "required_pages_per_second_upper_estimate": 0.0040028111912903
     },
     {
       "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": 19192.766291013566,
-      "required_pages_per_second_upper_estimate": 0.003997347071134101
+      "required_token_positions_per_second": 19121.14085838799,
+      "required_pages_per_second_upper_estimate": 0.0039824293824078745
     },
     {
       "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": 28789.14943652035,
-      "required_pages_per_second_upper_estimate": 0.0017258655502334845
+      "required_token_positions_per_second": 28681.71128758198,
+      "required_pages_per_second_upper_estimate": 0.0017194247972531838
     },
     {
       "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": 28789.14943652035,
-      "required_pages_per_second_upper_estimate": 0.002238229738603766
+      "required_token_positions_per_second": 28681.71128758198,
+      "required_pages_per_second_upper_estimate": 0.002229876895094282
     },
     {
       "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": 28789.14943652035,
-      "required_pages_per_second_upper_estimate": 0.36927629759136427
+      "required_token_positions_per_second": 28681.71128758198,
+      "required_pages_per_second_upper_estimate": 0.36789819637487947
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2724407482187985,
-    "producer_p95_seconds": 0.23576408999906562,
-    "gpu_update_seconds": 0.9603618217677264,
-    "producer_batches_per_second": 3.6705228808023063,
-    "gpu_batches_per_second": 1.0412742128371075,
-    "producer_headroom": 3.525030040647426,
+    "producer_batch_seconds": 0.25827292054690076,
+    "producer_p95_seconds": 0.22271191199979512,
+    "gpu_update_seconds": 0.9639592185690281,
+    "producer_batches_per_second": 3.8718732025117832,
+    "gpu_batches_per_second": 1.037388284417751,
+    "producer_headroom": 3.732327866691619,
     "sustainable": true,
-    "buffer_seconds": 60.502794771366766,
+    "buffer_seconds": 60.72943076984877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449478.4731698457,
-    "upload_bytes_per_second": 4959930.586827983,
+    "generation_bytes_per_second": 422231.52277767984,
+    "upload_bytes_per_second": 4952421.691545975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.44761701890495,
+    "estimated_full_shards_per_hour": 90.6010557412891,
     "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": 1.0780582049992518,
-      "bytes": 12237678,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.9231663379996462,
       "bytes": 6121395,
@@ -562,6 +556,12 @@
       "bytes": 7738259,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1072633740004676,
+      "bytes": 11735446,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -160,3 +160,4 @@
 {"step": 456200, "loss": 0.5912990570068359, "lr": 0.0003, "grad_norm": 0.1895758592389742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66773778104, "tokens_per_second": 152334.10924017965, "data_wait_s": 1.1068001185776666e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456300, "loss": 0.6288138628005981, "lr": 0.0003, "grad_norm": 0.17609634527369133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66788451869, "tokens_per_second": 152722.78976203222, "data_wait_s": 1.0080999345518649e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456400, "loss": 0.5894762873649597, "lr": 0.0003, "grad_norm": 0.1622674506318828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66803094828, "tokens_per_second": 151932.17241486724, "data_wait_s": 1.1503001587698236e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 456500, "loss": 0.6160978674888611, "lr": 0.0003, "grad_norm": 0.1599696314445684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66817832359, "tokens_per_second": 152691.77704402307, "data_wait_s": 1.2746000720653683e-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 sha256:72a8c18690958f0987f4f138fafedf7454b5ef7c0b69b36016c90684ecb4e021
-size 345630
+oid sha256:4eeb5431ab8cdff58f8c5b5be8a36d04e6223f2fdc5d059df703ca03a21de449
+size 347748