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": 74833.39606963248,
-      "required_pages_per_second_upper_estimate": 0.006486369107807393
+      "required_token_positions_per_second": 74481.04888003724,
+      "required_pages_per_second_upper_estimate": 0.0064558285464290655
     },
     {
       "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": 9354.17450870406,
-      "required_pages_per_second_upper_estimate": 0.003731520068894232
+      "required_token_positions_per_second": 9310.131110004655,
+      "required_pages_per_second_upper_estimate": 0.003713950498645546
     },
     {
       "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": 18708.34901740812,
-      "required_pages_per_second_upper_estimate": 0.0043278472894801405
+      "required_token_positions_per_second": 18620.26222000931,
+      "required_pages_per_second_upper_estimate": 0.004307469959176612
     },
     {
       "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": 28062.523526112178,
-      "required_pages_per_second_upper_estimate": 0.0017088604500580314
+      "required_token_positions_per_second": 27930.39333001396,
+      "required_pages_per_second_upper_estimate": 0.001700814414349213
     },
     {
       "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": 28062.523526112178,
-      "required_pages_per_second_upper_estimate": 0.0021603307430771075
+      "required_token_positions_per_second": 27930.39333001396,
+      "required_pages_per_second_upper_estimate": 0.0021501589948216744
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28062.523526112178,
-      "required_pages_per_second_upper_estimate": 0.348442622969718
+      "required_token_positions_per_second": 27930.39333001396,
+      "required_pages_per_second_upper_estimate": 0.3468020081454979
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10668362164071254,
-    "producer_p95_seconds": 0.3517403309997462,
-    "gpu_update_seconds": 0.9852285727002967,
-    "producer_batches_per_second": 9.373510053565528,
-    "gpu_batches_per_second": 1.0149928937395898,
-    "producer_headroom": 9.235049931266248,
+    "producer_batch_seconds": 0.10806918271882182,
+    "producer_p95_seconds": 0.3062480959997629,
+    "gpu_update_seconds": 0.9898893894304559,
+    "producer_batches_per_second": 9.253331753251388,
+    "gpu_batches_per_second": 1.010213879123769,
+    "producer_headroom": 9.159774919423466,
     "sustainable": true,
-    "buffer_seconds": 60.0989429347181,
+    "buffer_seconds": 60.38325275525781,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330078.32800613437,
-    "upload_bytes_per_second": 2808169.9437597515,
+    "generation_bytes_per_second": 330961.05595296423,
+    "upload_bytes_per_second": 2777055.609177954,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.82712535989783,
+    "estimated_full_shards_per_hour": 71.01653822843261,
     "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": 1.1035908939993533,
-      "bytes": 6360663,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9343603500019526,
       "bytes": 2109863,
@@ -562,6 +556,12 @@
       "bytes": 11538285,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.9019468660007988,
+      "bytes": 1029860,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 221400, "loss": 0.5887596476823092, "lr": 0.0003, "grad_norm": 0.15217444645913925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32406219146, "tokens_per_second": 145521.58387926136, "data_wait_s": 1.332699685008265e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221500, "loss": 0.5614271368831396, "lr": 0.0003, "grad_norm": 0.15897538511002124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32420766381, "tokens_per_second": 135064.96236603995, "data_wait_s": 1.1476997315185145e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221600, "loss": 0.558541651815176, "lr": 0.0003, "grad_norm": 0.14744477080345314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32435445354, "tokens_per_second": 148832.58113124082, "data_wait_s": 1.5354999050032347e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 221700, "loss": 0.6293789744377136, "lr": 0.0003, "grad_norm": 0.1453686250497285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32450090493, "tokens_per_second": 148181.2183745593, "data_wait_s": 1.0804997145896778e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dddc49cf0a55e6d8b487a9b30a36af09570285e5797192ee20bbddd1426b479f
-size 567468
+oid sha256:d7d263e14abd3f9ce6d54c5dbff6fd10466f002d9806b45c410ce8462a8f1048
+size 569586