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": 77372.15359242466,
-      "required_pages_per_second_upper_estimate": 0.0053527851136544215
+      "required_token_positions_per_second": 77484.9523633297,
+      "required_pages_per_second_upper_estimate": 0.005360588794354841
     },
     {
       "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": 9671.519199053082,
-      "required_pages_per_second_upper_estimate": 0.004049263124356524
+      "required_token_positions_per_second": 9685.619045416213,
+      "required_pages_per_second_upper_estimate": 0.004055166435590494
     },
     {
       "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": 19343.038398106164,
-      "required_pages_per_second_upper_estimate": 0.004028644788099529
+      "required_token_positions_per_second": 19371.238090832427,
+      "required_pages_per_second_upper_estimate": 0.004034518040418502
     },
     {
       "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": 29014.557597159244,
-      "required_pages_per_second_upper_estimate": 0.0017393784252854533
+      "required_token_positions_per_second": 29056.85713624864,
+      "required_pages_per_second_upper_estimate": 0.0017419142180662062
     },
     {
       "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": 29014.557597159244,
-      "required_pages_per_second_upper_estimate": 0.0022557542316277227
+      "required_token_positions_per_second": 29056.85713624864,
+      "required_pages_per_second_upper_estimate": 0.002259042834735914
     },
     {
       "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": 29014.557597159244,
-      "required_pages_per_second_upper_estimate": 0.37216759145161354
+      "required_token_positions_per_second": 29056.85713624864,
+      "required_pages_per_second_upper_estimate": 0.3727101645213458
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.26197179559379435,
-    "producer_p95_seconds": 0.21755887100152904,
-    "gpu_update_seconds": 0.95290096729605,
-    "producer_batches_per_second": 3.8172048167756585,
-    "gpu_batches_per_second": 1.0494269964250305,
-    "producer_headroom": 3.6374181622726662,
+    "producer_batch_seconds": 0.24702403241403204,
+    "producer_p95_seconds": 0.21369574500022281,
+    "gpu_update_seconds": 0.9515137810795414,
+    "producer_batches_per_second": 4.048189118392821,
+    "gpu_batches_per_second": 1.0509569276710302,
+    "producer_headroom": 3.8519077345670087,
     "sustainable": true,
-    "buffer_seconds": 60.03276093965115,
+    "buffer_seconds": 60.89688198909065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502483.6277302379,
-    "upload_bytes_per_second": 5795112.190165907,
+    "generation_bytes_per_second": 502385.60296099185,
+    "upload_bytes_per_second": 5836571.169068615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.82128929071763,
+    "estimated_full_shards_per_hour": 107.80025545713727,
     "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": 1.1097549460000664,
-      "bytes": 5080154,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1122178149998945,
       "bytes": 7607293,
@@ -562,6 +556,12 @@
       "bytes": 9643472,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1648727040010272,
+      "bytes": 8469052,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 538000, "loss": 0.5918370485305786, "lr": 0.0003, "grad_norm": 0.16774948382604338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78745674851, "tokens_per_second": 152304.7358001179, "data_wait_s": 1.3949000276625156e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538100, "loss": 0.5772097110748291, "lr": 0.0003, "grad_norm": 0.16119237883685197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78760368063, "tokens_per_second": 153216.76422369707, "data_wait_s": 9.27600012801122e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 538200, "loss": 0.5897762775421143, "lr": 0.0003, "grad_norm": 0.15242153421038196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78774895203, "tokens_per_second": 152190.3380791546, "data_wait_s": 9.842000508797355e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 538300, "loss": 0.5773850083351135, "lr": 0.0003, "grad_norm": 0.15849421215983392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78789289289, "tokens_per_second": 150842.4074431509, "data_wait_s": 1.0429999747429974e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2df610e6460a166b7247ec9203dfbd99fbd90b5bdac5233d479e8cf3b4c92915
-size 281307
+oid sha256:368ed36490d77da197b915a0d33db8a5c7d124d66f139d42ab61535d9c95d6f8
+size 283425