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": 76994.47146261457,
-      "required_pages_per_second_upper_estimate": 0.005326656187570882
+      "required_token_positions_per_second": 77349.15431738223,
+      "required_pages_per_second_upper_estimate": 0.005351193970441316
     },
     {
       "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": 9624.308932826822,
-      "required_pages_per_second_upper_estimate": 0.004029497171750054
+      "required_token_positions_per_second": 9668.644289672779,
+      "required_pages_per_second_upper_estimate": 0.004048059459833926
     },
     {
       "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": 19248.617865653643,
-      "required_pages_per_second_upper_estimate": 0.004008979481226557
+      "required_token_positions_per_second": 19337.288579345557,
+      "required_pages_per_second_upper_estimate": 0.004027447252484612
     },
     {
       "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": 28872.926798480465,
-      "required_pages_per_second_upper_estimate": 0.0017308878751623683
+      "required_token_positions_per_second": 29005.932869018336,
+      "required_pages_per_second_upper_estimate": 0.0017388613860026031
     },
     {
       "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": 28872.926798480465,
-      "required_pages_per_second_upper_estimate": 0.0022447430599984936
+      "required_token_positions_per_second": 29005.932869018336,
+      "required_pages_per_second_upper_estimate": 0.002255083696950937
     },
     {
       "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": 28872.926798480465,
-      "required_pages_per_second_upper_estimate": 0.3703509036374657
+      "required_token_positions_per_second": 29005.932869018336,
+      "required_pages_per_second_upper_estimate": 0.37205696269953353
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3767553764296565,
-    "producer_p95_seconds": 0.21884131499973591,
-    "gpu_update_seconds": 0.9575752466305241,
-    "producer_batches_per_second": 2.6542421490478945,
-    "gpu_batches_per_second": 1.0443043546904103,
-    "producer_headroom": 2.54163658049167,
+    "producer_batch_seconds": 0.12325689775769888,
+    "producer_p95_seconds": 0.2230575850007881,
+    "gpu_update_seconds": 0.9531843062882916,
+    "producer_batches_per_second": 8.113136207320599,
+    "gpu_batches_per_second": 1.0491150487926193,
+    "producer_headroom": 7.733314107597307,
     "sustainable": true,
-    "buffer_seconds": 60.327240537723014,
+    "buffer_seconds": 60.05061129616237,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373556.7056534131,
+    "generation_bytes_per_second": 387669.3345364263,
     "upload_bytes_per_second": 3703354.0885011475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.15657307817263,
+    "estimated_full_shards_per_hour": 83.18481471128074,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 942300, "loss": 0.5293458700180054, "lr": 3e-05, "grad_norm": 0.34303656299887136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137925881125, "tokens_per_second": 153226.535566637, "data_wait_s": 9.727002179715782e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 942400, "loss": 0.520447313785553, "lr": 3e-05, "grad_norm": 0.3352271217179194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137940681345, "tokens_per_second": 146237.1304412144, "data_wait_s": 1.642300048843026e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942500, "loss": 0.5312053561210632, "lr": 3e-05, "grad_norm": 0.3516962228627744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137955413024, "tokens_per_second": 153717.4198919749, "data_wait_s": 1.5871002688072622e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 942600, "loss": 0.47934281826019287, "lr": 3e-05, "grad_norm": 0.3301637447809434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137970109623, "tokens_per_second": 153375.35597927953, "data_wait_s": 1.4600998838432133e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77cb80ff6d1b69bc84a96f71791858aee617b94ea3e9c733172551c96ff6454d
-size 373164
+oid sha256:7fbf87ac19ac0910efdad132252f4a21971e3207b41d964a6bc2d374c927abee
+size 375282