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": 77044.0369419053,
-      "required_pages_per_second_upper_estimate": 0.0053300852424359715
+      "required_token_positions_per_second": 76812.92569325678,
+      "required_pages_per_second_upper_estimate": 0.005314096430002489
     },
     {
       "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": 9630.504617738163,
-      "required_pages_per_second_upper_estimate": 0.0040320911756418195
+      "required_token_positions_per_second": 9601.615711657098,
+      "required_pages_per_second_upper_estimate": 0.004019995993934636
     },
     {
       "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": 19261.009235476326,
-      "required_pages_per_second_upper_estimate": 0.004011560276778224
+      "required_token_positions_per_second": 19203.231423314195,
+      "required_pages_per_second_upper_estimate": 0.003999526682208225
     },
     {
       "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": 28891.51385321449,
-      "required_pages_per_second_upper_estimate": 0.0017320021407128837
+      "required_token_positions_per_second": 28804.847134971296,
+      "required_pages_per_second_upper_estimate": 0.001726806603286621
     },
     {
       "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": 28891.51385321449,
-      "required_pages_per_second_upper_estimate": 0.0022461881217482505
+      "required_token_positions_per_second": 28804.847134971296,
+      "required_pages_per_second_upper_estimate": 0.0022394501656114486
     },
     {
       "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": 28891.51385321449,
-      "required_pages_per_second_upper_estimate": 0.3705893184183693
+      "required_token_positions_per_second": 28804.847134971296,
+      "required_pages_per_second_upper_estimate": 0.3694776508122176
     }
   ],
   "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": 0.1168058204060003,
-    "producer_p95_seconds": 0.34213959900080226,
-    "gpu_update_seconds": 0.9569592005620663,
-    "producer_batches_per_second": 8.561217210958695,
-    "gpu_batches_per_second": 1.0449766295288805,
-    "producer_headroom": 8.192735578037235,
+    "producer_batch_seconds": 0.2671982125702925,
+    "producer_p95_seconds": 0.34721077600261196,
+    "gpu_update_seconds": 0.9598384560226743,
+    "producer_batches_per_second": 3.7425400057155227,
+    "gpu_batches_per_second": 1.0418419826016816,
+    "producer_headroom": 3.592233820689078,
     "sustainable": true,
-    "buffer_seconds": 60.288429635410175,
+    "buffer_seconds": 60.46982272942848,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 477392.7081342188,
-    "upload_bytes_per_second": 5353805.0970912315,
+    "generation_bytes_per_second": 505771.45201058703,
+    "upload_bytes_per_second": 5270847.12373085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.43736203212664,
+    "estimated_full_shards_per_hour": 108.52677984464844,
     "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": 1.1278007279979647,
-      "bytes": 7339680,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.694962838999345,
       "bytes": 1221938,
@@ -562,6 +556,12 @@
       "bytes": 6269271,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2005412010039436,
+      "bytes": 962816,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 434000, "loss": 0.591548502445221, "lr": 0.0003, "grad_norm": 0.1615272696927658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63524954041, "tokens_per_second": 151112.94545203622, "data_wait_s": 1.4924997230991721e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434100, "loss": 0.5987737774848938, "lr": 0.0003, "grad_norm": 0.19255513208487296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63539700982, "tokens_per_second": 153297.22819201264, "data_wait_s": 1.653799699852243e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434200, "loss": 0.5645946264266968, "lr": 0.0003, "grad_norm": 0.16347681515306023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63554366941, "tokens_per_second": 152190.84287311233, "data_wait_s": 1.9464001525193453e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 434300, "loss": 0.5650642514228821, "lr": 0.0003, "grad_norm": 0.13918712969755573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63568742841, "tokens_per_second": 148623.06905902762, "data_wait_s": 1.9120001525152475e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ab2881acba06ca085cb2d5b4332db7e144d944fe55e7fe534a0a41e1b16edc6
-size 791424
+oid sha256:d1708780bdd11275e11c5f56187ff593294adc688d57d29b9aed6632a725dfef
+size 793542