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": 77089.34462769584,
-      "required_pages_per_second_upper_estimate": 0.005333219733267276
+      "required_token_positions_per_second": 76911.29187502572,
+      "required_pages_per_second_upper_estimate": 0.005320901630698252
     },
     {
       "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": 9636.16807846198,
-      "required_pages_per_second_upper_estimate": 0.004034462348380373
+      "required_token_positions_per_second": 9613.911484378215,
+      "required_pages_per_second_upper_estimate": 0.004025143977208036
     },
     {
       "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": 19272.33615692396,
-      "required_pages_per_second_upper_estimate": 0.004013919375804761
+      "required_token_positions_per_second": 19227.82296875643,
+      "required_pages_per_second_upper_estimate": 0.004004648452601177
     },
     {
       "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": 28908.504235385943,
-      "required_pages_per_second_upper_estimate": 0.001733020688181251
+      "required_token_positions_per_second": 28841.734453134646,
+      "required_pages_per_second_upper_estimate": 0.00172901794168684
     },
     {
       "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": 28908.504235385943,
-      "required_pages_per_second_upper_estimate": 0.0022475090492292877
+      "required_token_positions_per_second": 28841.734453134646,
+      "required_pages_per_second_upper_estimate": 0.0022423179923484835
     },
     {
       "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": 28908.504235385943,
-      "required_pages_per_second_upper_estimate": 0.37080725279801363
+      "required_token_positions_per_second": 28841.734453134646,
+      "required_pages_per_second_upper_estimate": 0.36995080172310063
     }
   ],
   "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": 0.2537234906326944,
-    "producer_p95_seconds": 0.26276828499976546,
-    "gpu_update_seconds": 0.9563967673622145,
-    "producer_batches_per_second": 3.9412984485841753,
-    "gpu_batches_per_second": 1.0455911543470031,
-    "producer_headroom": 3.7694450954356165,
+    "producer_batch_seconds": 0.10330153316405699,
+    "producer_p95_seconds": 0.26625843299916596,
+    "gpu_update_seconds": 0.9586108645763186,
+    "producer_batches_per_second": 9.68039843524745,
+    "gpu_batches_per_second": 1.0431761593292335,
+    "producer_headroom": 9.279735113455802,
     "sustainable": true,
-    "buffer_seconds": 60.25299634381951,
+    "buffer_seconds": 60.39248446830807,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 495615.14072327246,
-    "upload_bytes_per_second": 4900356.331381153,
+    "generation_bytes_per_second": 495474.9940716568,
+    "upload_bytes_per_second": 4875682.583334658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.3474718692172,
+    "estimated_full_shards_per_hour": 106.31739966022757,
     "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": 1.5061208509996504,
-      "bytes": 11821724,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.1246318519988563,
       "bytes": 6660673,
@@ -562,6 +556,12 @@
       "bytes": 2347633,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.7378346169971337,
+      "bytes": 10782839,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -227,3 +227,4 @@
 {"step": 907700, "loss": 0.524348795413971, "lr": 3e-05, "grad_norm": 0.32118624783546756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132861030216, "tokens_per_second": 152002.71050669058, "data_wait_s": 1.8541999452281743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907800, "loss": 0.5214112997055054, "lr": 3e-05, "grad_norm": 0.31540677744893886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132875511591, "tokens_per_second": 150823.5107633387, "data_wait_s": 1.4773999282624573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907900, "loss": 0.5075257420539856, "lr": 3e-05, "grad_norm": 0.3304795518195391, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132890293246, "tokens_per_second": 152488.59864717792, "data_wait_s": 1.3494001905201003e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 908000, "loss": 0.49739935994148254, "lr": 3e-05, "grad_norm": 0.32429283499638284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132905040804, "tokens_per_second": 153394.06379499636, "data_wait_s": 2.1040999854449183e-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 sha256:4d4642b66eedf2b80ef8d7f99265a7280c12800c8feb8224a595d8ab4724e31e
-size 488319
+oid sha256:0547178d64d2e416e862563f2fa3a4de0388f20dd099436842f949e7acf2e908
+size 490437