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": 75720.26183877254,
-      "required_pages_per_second_upper_estimate": 0.005238503409219827
+      "required_token_positions_per_second": 75991.67325524974,
+      "required_pages_per_second_upper_estimate": 0.0052572802807729704
     },
     {
       "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": 9465.032729846567,
-      "required_pages_per_second_upper_estimate": 0.00396281155162756
+      "required_token_positions_per_second": 9498.959156906218,
+      "required_pages_per_second_upper_estimate": 0.003977015838173076
     },
     {
       "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": 18930.065459693134,
-      "required_pages_per_second_upper_estimate": 0.003942633415856858
+      "required_token_positions_per_second": 18997.918313812435,
+      "required_pages_per_second_upper_estimate": 0.003956765375969813
     },
     {
       "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": 28395.0981895397,
-      "required_pages_per_second_upper_estimate": 0.0017022427796584102
+      "required_token_positions_per_second": 28496.877470718653,
+      "required_pages_per_second_upper_estimate": 0.0017083442921571243
     },
     {
       "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": 28395.0981895397,
-      "required_pages_per_second_upper_estimate": 0.0022075939873992807
+      "required_token_positions_per_second": 28496.877470718653,
+      "required_pages_per_second_upper_estimate": 0.0022155068788311987
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28395.0981895397,
-      "required_pages_per_second_upper_estimate": 0.3642218312943613
+      "required_token_positions_per_second": 28496.877470718653,
+      "required_pages_per_second_upper_estimate": 0.3655273466312471
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07481331355481302,
-    "producer_p95_seconds": 0.26225396399968304,
-    "gpu_update_seconds": 0.9736891845010446,
-    "producer_batches_per_second": 13.366604852588651,
-    "gpu_batches_per_second": 1.0270217805823099,
-    "producer_headroom": 13.014918578464748,
+    "producer_batch_seconds": 0.3135869689531603,
+    "producer_p95_seconds": 0.2939966670019203,
+    "gpu_update_seconds": 0.970211561895127,
+    "producer_batches_per_second": 3.1889080191637924,
+    "gpu_batches_per_second": 1.0307030335184697,
+    "producer_headroom": 3.0939154300127987,
     "sustainable": true,
-    "buffer_seconds": 60.368729439064765,
+    "buffer_seconds": 60.153116837497876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398258.3386686484,
+    "generation_bytes_per_second": 379780.2179360584,
     "upload_bytes_per_second": 4843504.338466388,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.45696849865521,
+    "estimated_full_shards_per_hour": 81.49199393807712,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 865900, "loss": 0.555191159248352, "lr": 3e-05, "grad_norm": 0.32342886147925076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126743758706, "tokens_per_second": 152578.89325919474, "data_wait_s": 9.358998795505613e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 866000, "loss": 0.5319803953170776, "lr": 3e-05, "grad_norm": 0.3136856848495784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126758284351, "tokens_per_second": 150123.62397090017, "data_wait_s": 1.3885000953450799e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 866100, "loss": 0.49722954630851746, "lr": 3e-05, "grad_norm": 0.31966955954025744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126772931548, "tokens_per_second": 151866.74268190228, "data_wait_s": 6.505000783363357e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 866200, "loss": 0.5219362378120422, "lr": 3e-05, "grad_norm": 0.3036566328829848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126787579553, "tokens_per_second": 145926.98003303452, "data_wait_s": 1.4743000065209344e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b33af9e2bde2f93a4413cabc01e0bbe79df079f09f37e9bdea787113cc31df54
-size 503928
+oid sha256:3b1c0ea8881a4d07f91ddf30ecc12a58b114504d8e3e92837640c17c13ed5e9b
+size 506046