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": 76942.26587933519,
-      "required_pages_per_second_upper_estimate": 0.005323044484185972
+      "required_token_positions_per_second": 77650.02462904147,
+      "required_pages_per_second_upper_estimate": 0.005372008876717207
     },
     {
       "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": 9617.783234916898,
-      "required_pages_per_second_upper_estimate": 0.004026764998307238
+      "required_token_positions_per_second": 9706.253078630183,
+      "required_pages_per_second_upper_estimate": 0.004063805474409572
     },
     {
       "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": 19235.566469833797,
-      "required_pages_per_second_upper_estimate": 0.004006261219665737
+      "required_token_positions_per_second": 19412.506157260366,
+      "required_pages_per_second_upper_estimate": 0.004043113090343345
     },
     {
       "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": 28853.349704750693,
-      "required_pages_per_second_upper_estimate": 0.0017297142582857598
+      "required_token_positions_per_second": 29118.75923589055,
+      "required_pages_per_second_upper_estimate": 0.0017456251544207054
     },
     {
       "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": 28853.349704750693,
-      "required_pages_per_second_upper_estimate": 0.0022432210270715387
+      "required_token_positions_per_second": 29118.75923589055,
+      "required_pages_per_second_upper_estimate": 0.002263855450704511
     },
     {
       "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": 28853.349704750693,
-      "required_pages_per_second_upper_estimate": 0.37009978969934576
+      "required_token_positions_per_second": 29118.75923589055,
+      "required_pages_per_second_upper_estimate": 0.373504178190256
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5607584895236073,
-    "producer_p95_seconds": 0.3259918949988787,
-    "gpu_update_seconds": 0.9582249646198884,
-    "producer_batches_per_second": 1.7832989043992016,
-    "gpu_batches_per_second": 1.043596271149837,
-    "producer_headroom": 1.7088015295746106,
+    "producer_batch_seconds": 0.1183229249611486,
+    "producer_p95_seconds": 0.31304708900279365,
+    "gpu_update_seconds": 0.9494910059877224,
+    "producer_batches_per_second": 8.451447598412146,
+    "gpu_batches_per_second": 1.0531958635666432,
+    "producer_headroom": 8.02457348226887,
     "sustainable": true,
-    "buffer_seconds": 60.36817277105297,
+    "buffer_seconds": 60.767424383214234,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247002.13960591122,
-    "upload_bytes_per_second": 5096453.290684326,
+    "generation_bytes_per_second": 248576.8474397672,
+    "upload_bytes_per_second": 5145087.764592436,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.00090924389841,
+    "estimated_full_shards_per_hour": 53.33880488772166,
     "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": 1.323775843000476,
-      "bytes": 7235009,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.370845877998363,
       "bytes": 11121979,
@@ -562,6 +556,12 @@
       "bytes": 763221,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.1390506280004047,
+      "bytes": 10406852,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -400,3 +400,4 @@
 {"step": 925000, "loss": 0.5517297387123108, "lr": 3e-05, "grad_norm": 0.34564684448204536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135393509500, "tokens_per_second": 150325.96381885314, "data_wait_s": 9.957002475857735e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 925100, "loss": 0.5094435214996338, "lr": 3e-05, "grad_norm": 0.33234581361357024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135408062366, "tokens_per_second": 138679.39196748857, "data_wait_s": 1.405299553880468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925200, "loss": 0.5638466477394104, "lr": 3e-05, "grad_norm": 0.33840653702274737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135422770685, "tokens_per_second": 154307.70573229526, "data_wait_s": 1.7597005353309214e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 925300, "loss": 0.5204133987426758, "lr": 3e-05, "grad_norm": 0.3171108751895482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135437541266, "tokens_per_second": 154289.62706155606, "data_wait_s": 1.407499803462997e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5784f60e0b8b2af2c2d2084da64dd48c415846098fd19019dc1ea7213961a723
-size 857865
+oid sha256:b54b1f4357aada9ccf238d590c21f14d6849d757892ce68244b1158dccce8290
+size 859983