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": 76640.9884690638,
-      "required_pages_per_second_upper_estimate": 0.005302201413883494
+      "required_token_positions_per_second": 76254.89113553984,
+      "required_pages_per_second_upper_estimate": 0.005275490304481066
     },
     {
       "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": 9580.123558632975,
-      "required_pages_per_second_upper_estimate": 0.004010997678270627
+      "required_token_positions_per_second": 9531.86139194248,
+      "required_pages_per_second_upper_estimate": 0.003990791316905961
     },
     {
       "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": 19160.24711726595,
-      "required_pages_per_second_upper_estimate": 0.003990574184830755
+      "required_token_positions_per_second": 19063.72278388496,
+      "required_pages_per_second_upper_estimate": 0.003970470711705345
     },
     {
       "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": 28740.370675898925,
-      "required_pages_per_second_upper_estimate": 0.0017229413380150849
+      "required_token_positions_per_second": 28595.58417582744,
+      "required_pages_per_second_upper_estimate": 0.0017142616084119841
     },
     {
       "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": 28740.370675898925,
-      "required_pages_per_second_upper_estimate": 0.002234437404520543
+      "required_token_positions_per_second": 28595.58417582744,
+      "required_pages_per_second_upper_estimate": 0.0022231808909884947
     },
     {
       "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": 28740.370675898925,
-      "required_pages_per_second_upper_estimate": 0.3686506160246652
+      "required_token_positions_per_second": 28595.58417582744,
+      "required_pages_per_second_upper_estimate": 0.36679345026137994
     }
   ],
   "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": 0.1369418721486113,
-    "producer_p95_seconds": 0.31217062499490567,
-    "gpu_update_seconds": 0.961991768017976,
-    "producer_batches_per_second": 7.302368401351965,
-    "gpu_batches_per_second": 1.0395099347475016,
-    "producer_headroom": 7.024818289135179,
+    "producer_batch_seconds": 0.4203665099685736,
+    "producer_p95_seconds": 0.3029203490004875,
+    "gpu_update_seconds": 0.966862569758989,
+    "producer_batches_per_second": 2.3788764715694395,
+    "gpu_batches_per_second": 1.0342731545076476,
+    "producer_headroom": 2.300046618440825,
     "sustainable": true,
-    "buffer_seconds": 60.60548138513249,
+    "buffer_seconds": 60.91234189481631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333966.07981031085,
-    "upload_bytes_per_second": 5105576.1864239555,
+    "generation_bytes_per_second": 320672.3926296532,
+    "upload_bytes_per_second": 5125620.756398993,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.661346394844,
+    "estimated_full_shards_per_hour": 68.80883058707425,
     "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": 1.182239486999606,
-      "bytes": 5356316,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.588657971999055,
       "bytes": 13335657,
@@ -562,6 +556,12 @@
       "bytes": 6281809,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1541477030041278,
+      "bytes": 6906912,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 756700, "loss": 0.499985933303833, "lr": 5e-05, "grad_norm": 0.23385300021185157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110759322182, "tokens_per_second": 149120.42494103522, "data_wait_s": 1.2472002708818763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756800, "loss": 0.5425806045532227, "lr": 5e-05, "grad_norm": 0.24550936957946895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110773884834, "tokens_per_second": 150261.21545031416, "data_wait_s": 1.2894997780676931e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756900, "loss": 0.5776959657669067, "lr": 5e-05, "grad_norm": 0.2595629228742037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110788617501, "tokens_per_second": 152651.10328770976, "data_wait_s": 1.20049953693524e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 757000, "loss": 0.5287701487541199, "lr": 5e-05, "grad_norm": 0.24790329972095032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110803297550, "tokens_per_second": 152267.81768152904, "data_wait_s": 1.304799661738798e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a2ea3ce22efe168a8244c977cad3c4e04d58a83ecece6a0b5897eb455cc10093
-size 787188
+oid sha256:a043654701c10f8e7b6b2721cdc274df21155edd494d11feea426c1bb138a580
+size 789306