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": 76386.45520285396,
-      "required_pages_per_second_upper_estimate": 0.005284592211928551
+      "required_token_positions_per_second": 77389.27735885547,
+      "required_pages_per_second_upper_estimate": 0.005353969775548717
     },
     {
       "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": 9548.306900356745,
-      "required_pages_per_second_upper_estimate": 0.003997676707857747
+      "required_token_positions_per_second": 9673.659669856934,
+      "required_pages_per_second_upper_estimate": 0.004050159294783984
     },
     {
       "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": 19096.61380071349,
-      "required_pages_per_second_upper_estimate": 0.003977321043116284
+      "required_token_positions_per_second": 19347.319339713868,
+      "required_pages_per_second_upper_estimate": 0.004029536395340392
     },
     {
       "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": 28644.920701070234,
-      "required_pages_per_second_upper_estimate": 0.0017172192577677775
+      "required_token_positions_per_second": 29020.9790095708,
+      "required_pages_per_second_upper_estimate": 0.0017397633791545971
     },
     {
       "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": 28644.920701070234,
-      "required_pages_per_second_upper_estimate": 0.0022270165888176817
+      "required_token_positions_per_second": 29020.9790095708,
+      "required_pages_per_second_upper_estimate": 0.002256253468197913
     },
     {
       "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": 28644.920701070234,
-      "required_pages_per_second_upper_estimate": 0.3674262862337609
+      "required_token_positions_per_second": 29020.9790095708,
+      "required_pages_per_second_upper_estimate": 0.37224995843525355
     }
   ],
   "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": 0.10194944192977573,
-    "producer_p95_seconds": 0.24927621299866587,
-    "gpu_update_seconds": 0.9651972958321722,
-    "producer_batches_per_second": 9.80878346238339,
-    "gpu_batches_per_second": 1.036057606375515,
-    "producer_headroom": 9.46741127329578,
+    "producer_batch_seconds": 0.4578429652029854,
+    "producer_p95_seconds": 0.2389157139987219,
+    "gpu_update_seconds": 0.9526901208564326,
+    "producer_batches_per_second": 2.1841549963679108,
+    "gpu_batches_per_second": 1.049659252371629,
+    "producer_headroom": 2.080822887458926,
     "sustainable": true,
-    "buffer_seconds": 60.80742963742685,
+    "buffer_seconds": 60.01947761395525,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333002.43152754265,
-    "upload_bytes_per_second": 5629672.508276998,
+    "generation_bytes_per_second": 299641.3044262727,
+    "upload_bytes_per_second": 5592687.288932413,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.45456990594587,
+    "estimated_full_shards_per_hour": 64.2960486373056,
     "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": 1.312004507999518,
-      "bytes": 5834009,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.3078732329995546,
       "bytes": 7188440,
@@ -562,6 +556,12 @@
       "bytes": 5790944,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2694779230005224,
+      "bytes": 2738639,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 707400, "loss": 0.5188702344894409, "lr": 8e-05, "grad_norm": 0.19953598797003128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103543182303, "tokens_per_second": 151694.7663148193, "data_wait_s": 1.6600999515503645e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707500, "loss": 0.5464622974395752, "lr": 8e-05, "grad_norm": 0.19706057669124938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103557788498, "tokens_per_second": 152070.82317731014, "data_wait_s": 1.650600097491406e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 707600, "loss": 0.5644626617431641, "lr": 8e-05, "grad_norm": 0.23308038512785945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103572462469, "tokens_per_second": 134158.50991644702, "data_wait_s": 1.487299959990196e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 707700, "loss": 0.5445774793624878, "lr": 8e-05, "grad_norm": 0.23317238347965258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103587149186, "tokens_per_second": 153324.85007495372, "data_wait_s": 1.4682002074550837e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48ce9479b2533a8a065b79d232a4837da3c063bd3877ee3ab42cc8dc0727b493
-size 481965
+oid sha256:09d4698f0f5599956a6a742838fcdbdd90ab2eb02515a915ed356161b260d6f8
+size 484083