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": 74895.19768756824,
-      "required_pages_per_second_upper_estimate": 0.006491725915415297
+      "required_token_positions_per_second": 74881.78874324774,
+      "required_pages_per_second_upper_estimate": 0.006490563662106254
     },
     {
       "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": 9361.89971094603,
-      "required_pages_per_second_upper_estimate": 0.00373460176757062
+      "required_token_positions_per_second": 9360.223592905968,
+      "required_pages_per_second_upper_estimate": 0.0037339331390242414
     },
     {
       "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": 18723.79942189206,
-      "required_pages_per_second_upper_estimate": 0.004331421468639669
+      "required_token_positions_per_second": 18720.447185811936,
+      "required_pages_per_second_upper_estimate": 0.004330645987819871
     },
     {
       "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": 28085.69913283809,
-      "required_pages_per_second_upper_estimate": 0.001710271722914628
+      "required_token_positions_per_second": 28080.670778717904,
+      "required_pages_per_second_upper_estimate": 0.001709965522530444
     },
     {
       "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": 28085.69913283809,
-      "required_pages_per_second_upper_estimate": 0.0021621148654370533
+      "required_token_positions_per_second": 28080.670778717904,
+      "required_pages_per_second_upper_estimate": 0.0021617277688175066
     },
     {
       "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": 28085.69913283809,
-      "required_pages_per_second_upper_estimate": 0.3487303864414876
+      "required_token_positions_per_second": 28080.670778717904,
+      "required_pages_per_second_upper_estimate": 0.3486679511121336
     }
   ],
   "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": 0.06324667857802524,
-    "producer_p95_seconds": 0.21876399199936714,
-    "gpu_update_seconds": 0.9844155870655779,
-    "producer_batches_per_second": 15.81110696218355,
-    "gpu_batches_per_second": 1.0158311318300814,
-    "producer_headroom": 15.564700142334564,
+    "producer_batch_seconds": 0.09122895850774171,
+    "producer_p95_seconds": 0.23852974600049492,
+    "gpu_update_seconds": 0.9845918645559628,
+    "producer_batches_per_second": 10.961431724720827,
+    "gpu_batches_per_second": 1.015649261383026,
+    "producer_headroom": 10.792536500045763,
     "sustainable": true,
-    "buffer_seconds": 60.04935081100025,
+    "buffer_seconds": 60.060103737913735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345581.92676408176,
-    "upload_bytes_per_second": 5357803.870381837,
+    "generation_bytes_per_second": 346212.69621597667,
+    "upload_bytes_per_second": 5206602.43684275,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.15383674804535,
+    "estimated_full_shards_per_hour": 74.28918518886066,
     "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": 1.256968259000132,
-      "bytes": 11567319,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1466406099998494,
       "bytes": 5899615,
@@ -562,6 +556,12 @@
       "bytes": 7451679,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.598954668999795,
+      "bytes": 1262837,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -146,3 +146,4 @@
 {"step": 209600, "loss": 0.616977671161294, "lr": 0.0003, "grad_norm": 0.1456166156245679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30677633997, "tokens_per_second": 148708.33548059704, "data_wait_s": 1.6396999853895977e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209700, "loss": 0.574382184073329, "lr": 0.0003, "grad_norm": 0.1353080839383903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30692338469, "tokens_per_second": 149167.40043659296, "data_wait_s": 1.6989000869216397e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209800, "loss": 0.5944355335086584, "lr": 0.0003, "grad_norm": 0.1378234884751291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30706866861, "tokens_per_second": 147099.06887911726, "data_wait_s": 1.7036998542607762e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 209900, "loss": 0.6171294096857309, "lr": 0.0003, "grad_norm": 0.14364944481939032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30721359198, "tokens_per_second": 146652.94399301035, "data_wait_s": 1.898600021377206e-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:f064985079ef2c6d9ed80c7db295f0f8b4f95cb245545c0ced5da302445cb51d
-size 315195
+oid sha256:5e79d6a506db98868b8dfa0d4e870892ae466b20041f646410a862ccd9976011
+size 317313