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": 76038.62316351659,
-      "required_pages_per_second_upper_estimate": 0.006590835137685125
+      "required_token_positions_per_second": 75005.50320983281,
+      "required_pages_per_second_upper_estimate": 0.006501286918518399
     },
     {
       "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": 9504.827895439574,
-      "required_pages_per_second_upper_estimate": 0.0037916179573318867
+      "required_token_positions_per_second": 9375.687901229101,
+      "required_pages_per_second_upper_estimate": 0.003740102082826353
     },
     {
       "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": 19009.655790879147,
-      "required_pages_per_second_upper_estimate": 0.004397549308704563
+      "required_token_positions_per_second": 18751.375802458202,
+      "required_pages_per_second_upper_estimate": 0.004337800778955923
     },
     {
       "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": 28514.48368631872,
-      "required_pages_per_second_upper_estimate": 0.001736382452562909
+      "required_token_positions_per_second": 28127.0637036873,
+      "required_pages_per_second_upper_estimate": 0.0017127906082562143
     },
     {
       "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": 28514.48368631872,
-      "required_pages_per_second_upper_estimate": 0.0021951238873156045
+      "required_token_positions_per_second": 28127.0637036873,
+      "required_pages_per_second_upper_estimate": 0.0021652992246054864
     },
     {
       "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": 28514.48368631872,
-      "required_pages_per_second_upper_estimate": 0.3540544555461306
+      "required_token_positions_per_second": 28127.0637036873,
+      "required_pages_per_second_upper_estimate": 0.34924399597312167
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10947339260135891,
-    "producer_p95_seconds": 0.37057545100105926,
-    "gpu_update_seconds": 0.9696125065475248,
-    "producer_batches_per_second": 9.13463971689854,
-    "gpu_batches_per_second": 1.0313398324044676,
-    "producer_headroom": 8.857060912310567,
+    "producer_batch_seconds": 0.2612360989295155,
+    "producer_p95_seconds": 0.35333787000126904,
+    "gpu_update_seconds": 0.9829678736204341,
+    "producer_batches_per_second": 3.8279548810358386,
+    "gpu_batches_per_second": 1.017327246227116,
+    "producer_headroom": 3.7627566697267603,
     "sustainable": true,
-    "buffer_seconds": 60.11597540594654,
+    "buffer_seconds": 60.944008164466915,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 553240.2388014272,
+    "generation_bytes_per_second": 525314.4041183096,
     "upload_bytes_per_second": 5258435.183261904,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.71247647316085,
+    "estimated_full_shards_per_hour": 112.72024243032423,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 146400, "loss": 0.5712147578597069, "lr": 0.0003, "grad_norm": 0.1339102732111261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21425080478, "tokens_per_second": 149677.5122500821, "data_wait_s": 1.1968000762863085e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 146500, "loss": 0.604755088686943, "lr": 0.0003, "grad_norm": 0.14296582557265414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21439675555, "tokens_per_second": 149319.34816546435, "data_wait_s": 9.305997082265094e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 146600, "loss": 0.614015381783247, "lr": 0.0003, "grad_norm": 0.1371328897020744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21454369327, "tokens_per_second": 150034.3524112255, "data_wait_s": 9.960000170394778e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 146700, "loss": 0.6494546383619308, "lr": 0.0003, "grad_norm": 0.15985365485145217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21469065047, "tokens_per_second": 150763.3991742238, "data_wait_s": 1.048400008585304e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fef5fca4d1231793c244ba068bfec07714d07b3ba00577a9810830d0b2415a4
-size 674151
+oid sha256:35a395507f2fd549e45bc3d0ce3ddb5c623c44dfdb73f40c129bb474019006d7
+size 676269