Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75407.0958612379,
-      "required_pages_per_second_upper_estimate": 0.006536095951714923
+      "required_token_positions_per_second": 75515.9295437769,
+      "required_pages_per_second_upper_estimate": 0.006545529379480957
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9425.886982654738,
-      "required_pages_per_second_upper_estimate": 0.0037601272469501906
+      "required_token_positions_per_second": 9439.491192972113,
+      "required_pages_per_second_upper_estimate": 0.003765554169846862
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18851.773965309476,
-      "required_pages_per_second_upper_estimate": 0.004361026126984248
+      "required_token_positions_per_second": 18878.982385944226,
+      "required_pages_per_second_upper_estimate": 0.004367320316246249
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28277.660947964214,
-      "required_pages_per_second_upper_estimate": 0.0017219611903100014
+      "required_token_positions_per_second": 28318.473578916335,
+      "required_pages_per_second_upper_estimate": 0.0017244464654076596
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28277.660947964214,
-      "required_pages_per_second_upper_estimate": 0.002176892617349777
+      "required_token_positions_per_second": 28318.473578916335,
+      "required_pages_per_second_upper_estimate": 0.0021800344866570644
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28277.660947964214,
-      "required_pages_per_second_upper_estimate": 0.3511139097304868
+      "required_token_positions_per_second": 28318.473578916335,
+      "required_pages_per_second_upper_estimate": 0.351620666015823
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.17497796114065522,
-    "producer_p95_seconds": 0.29304440899795736,
-    "gpu_update_seconds": 0.9777329196667681,
-    "producer_batches_per_second": 5.715005441148984,
-    "gpu_batches_per_second": 1.0227741951665297,
-    "producer_headroom": 5.587748955886061,
+    "producer_batch_seconds": 0.08516380454679506,
+    "producer_p95_seconds": 0.23862360199927934,
+    "gpu_update_seconds": 0.9763238093660699,
+    "producer_batches_per_second": 11.742077580041986,
+    "gpu_batches_per_second": 1.0242503464596475,
+    "producer_headroom": 11.464069912818514,
     "sustainable": true,
-    "buffer_seconds": 60.619441019339625,
+    "buffer_seconds": 60.532076180696336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370439.1585049232,
+    "generation_bytes_per_second": 370025.6959069958,
     "upload_bytes_per_second": 5918507.663151537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.48762003289006,
+    "estimated_full_shards_per_hour": 79.39890058429151,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -182,3 +182,4 @@
 {"step": 333200, "loss": 0.570095119997859, "lr": 0.0003, "grad_norm": 0.16794487178184855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48772284324, "tokens_per_second": 148634.77690898417, "data_wait_s": 1.9146998965879902e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333300, "loss": 0.5874333307147026, "lr": 0.0003, "grad_norm": 0.13781476407410762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48787016031, "tokens_per_second": 149138.41870287724, "data_wait_s": 1.671800055191852e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333400, "loss": 0.672693058848381, "lr": 0.0003, "grad_norm": 0.17768767211908643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48801698217, "tokens_per_second": 149498.3728157396, "data_wait_s": 1.829400207498111e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 333500, "loss": 0.6238420605659485, "lr": 0.0003, "grad_norm": 0.20893165506957004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48816459158, "tokens_per_second": 149701.1962721406, "data_wait_s": 1.8580001778900623e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2aa73aafd5126e840203dc7785d847422b404bb5fee03344b49c6579404a5ad5
-size 392226
+oid sha256:b2f32897e114f4a0410a6f02d29cf022ce22ace4715113b4dda8da5b3978e6c6
+size 394344