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": 76404.5823321046,
-      "required_pages_per_second_upper_estimate": 0.005285846288790846
+      "required_token_positions_per_second": 75810.75598818513,
+      "required_pages_per_second_upper_estimate": 0.005244764004451546
     },
     {
       "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": 9550.572791513076,
-      "required_pages_per_second_upper_estimate": 0.003998625389167714
+      "required_token_positions_per_second": 9476.344498523142,
+      "required_pages_per_second_upper_estimate": 0.0039675475529558505
     },
     {
       "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": 19101.14558302615,
-      "required_pages_per_second_upper_estimate": 0.003978264893860875
+      "required_token_positions_per_second": 18952.688997046283,
+      "required_pages_per_second_upper_estimate": 0.0039473453020647175
     },
     {
       "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": 28651.718374539225,
-      "required_pages_per_second_upper_estimate": 0.001717626767912797
+      "required_token_positions_per_second": 28429.033495569423,
+      "required_pages_per_second_upper_estimate": 0.0017042771494386799
     },
     {
       "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": 28651.718374539225,
-      "required_pages_per_second_upper_estimate": 0.0022275450780300902
+      "required_token_positions_per_second": 28429.033495569423,
+      "required_pages_per_second_upper_estimate": 0.0022102323081774553
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28651.718374539225,
-      "required_pages_per_second_upper_estimate": 0.367513479490248
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28429.033495569423,
+      "required_pages_per_second_upper_estimate": 0.3646571169632178
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07375040761729679,
-    "producer_p95_seconds": 0.2555529290002596,
-    "gpu_update_seconds": 0.9649683009787239,
-    "producer_batches_per_second": 13.559247091747173,
-    "gpu_batches_per_second": 1.036303471301332,
-    "producer_headroom": 13.084243628673972,
+    "producer_batch_seconds": 0.12072705212511892,
+    "producer_p95_seconds": 0.2712036260018067,
+    "gpu_update_seconds": 0.972526906491874,
+    "producer_batches_per_second": 8.283147665724675,
+    "gpu_batches_per_second": 1.0282491860376672,
+    "producer_headroom": 8.055583975362605,
     "sustainable": true,
-    "buffer_seconds": 60.7930029616596,
+    "buffer_seconds": 60.296668202496186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379418.5529614432,
+    "generation_bytes_per_second": 379882.6176367886,
     "upload_bytes_per_second": 4881374.903585045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.4143890536544,
+    "estimated_full_shards_per_hour": 81.51396653011076,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 782900, "loss": 0.5246395468711853, "lr": 3e-05, "grad_norm": 0.25183891041606093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114594406316, "tokens_per_second": 152931.21247063638, "data_wait_s": 8.129998604999855e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 783000, "loss": 0.5296126008033752, "lr": 3e-05, "grad_norm": 0.26037759137340166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114609159455, "tokens_per_second": 152681.0293511491, "data_wait_s": 1.2995998986298218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783100, "loss": 0.5236709117889404, "lr": 3e-05, "grad_norm": 0.25507218812778376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114623872966, "tokens_per_second": 152265.31835006358, "data_wait_s": 1.296800110139884e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 783200, "loss": 0.5158957839012146, "lr": 3e-05, "grad_norm": 0.24981190451559307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114638666763, "tokens_per_second": 153343.64682076938, "data_wait_s": 1.4785997336730361e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa1f0adfb1d5d6027ab141f3c84303f5fb9bfdd1ee78c57f503e29b7d3aa05f9
-size 492555
+oid sha256:da234903c7b62d858a9c62b53bdc8e83e2d2e89f3dae78fa56050de422e79745
+size 494673