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": 76196.42757821929,
-      "required_pages_per_second_upper_estimate": 0.0052714456599314875
+      "required_token_positions_per_second": 76367.54238620421,
+      "required_pages_per_second_upper_estimate": 0.005283283779388941
     },
     {
       "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": 9524.55344727741,
-      "required_pages_per_second_upper_estimate": 0.0039877316330819345
+      "required_token_positions_per_second": 9545.942798275526,
+      "required_pages_per_second_upper_estimate": 0.003996686907684405
     },
     {
       "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": 19049.10689455482,
-      "required_pages_per_second_upper_estimate": 0.003967426607404799
+      "required_token_positions_per_second": 19091.885596551052,
+      "required_pages_per_second_upper_estimate": 0.003976336282880377
     },
     {
       "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": 28573.66034183223,
-      "required_pages_per_second_upper_estimate": 0.0017129473080397274
+      "required_token_positions_per_second": 28637.828394826574,
+      "required_pages_per_second_upper_estimate": 0.0017167940848377948
     },
     {
       "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": 28573.66034183223,
-      "required_pages_per_second_upper_estimate": 0.0022214764093281218
+      "required_token_positions_per_second": 28637.828394826574,
+      "required_pages_per_second_upper_estimate": 0.0022264651932029957
     },
     {
       "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": 28573.66034183223,
-      "required_pages_per_second_upper_estimate": 0.3665122348588683
+      "required_token_positions_per_second": 28637.828394826574,
+      "required_pages_per_second_upper_estimate": 0.3673353137443924
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09997676446874948,
-    "producer_p95_seconds": 0.2777497159986524,
-    "gpu_update_seconds": 0.9676044185183705,
-    "producer_batches_per_second": 10.00232409314044,
-    "gpu_batches_per_second": 1.0334801917618717,
-    "producer_headroom": 9.678292987975443,
+    "producer_batch_seconds": 0.07650065592190458,
+    "producer_p95_seconds": 0.28155908600092516,
+    "gpu_update_seconds": 0.9654363319320194,
+    "producer_batches_per_second": 13.071783345502899,
+    "gpu_batches_per_second": 1.035801084882327,
+    "producer_headroom": 12.61997456489238,
     "sustainable": true,
-    "buffer_seconds": 60.959078366657344,
+    "buffer_seconds": 60.82248891171722,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292214.6060177001,
-    "upload_bytes_per_second": 4908015.375771855,
+    "generation_bytes_per_second": 323104.31627050944,
+    "upload_bytes_per_second": 4909392.785882454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.70245204351666,
+    "estimated_full_shards_per_hour": 69.33066478811705,
     "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": 0.93341918599981,
-      "bytes": 1061744,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.5622991370000818,
       "bytes": 16646339,
@@ -562,6 +556,12 @@
       "bytes": 949705,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.9317821920012648,
+      "bytes": 6082541,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 867800, "loss": 0.4708053767681122, "lr": 3e-05, "grad_norm": 0.291191512431321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127022301013, "tokens_per_second": 152130.24573968275, "data_wait_s": 1.0878000466618687e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 867900, "loss": 0.4677457809448242, "lr": 3e-05, "grad_norm": 0.3257396143011411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127036817773, "tokens_per_second": 150672.52302423256, "data_wait_s": 9.336003131465986e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 868000, "loss": 0.5193307399749756, "lr": 3e-05, "grad_norm": 0.30677056004051867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127051330628, "tokens_per_second": 150989.08953530111, "data_wait_s": 1.0551000741543248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 868100, "loss": 0.5418099761009216, "lr": 3e-05, "grad_norm": 0.3158439129922565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127065902252, "tokens_per_second": 150349.03504273325, "data_wait_s": 1.3627002772409469e-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 sha256:780d32bb8015f4165d2930828f01c2a18b81ec1ebecea986c20919c9ba7a0f65
-size 544170
+oid sha256:bb80fd42116c78be5013922b48f7749de4f2d91ebcc280fb34f9323b1a3ed033
+size 546288