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": 75961.66188234878,
-      "required_pages_per_second_upper_estimate": 0.006584164328890083
+      "required_token_positions_per_second": 75479.26315642834,
+      "required_pages_per_second_upper_estimate": 0.006542351229955692
     },
     {
       "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": 9495.207735293598,
-      "required_pages_per_second_upper_estimate": 0.003787780331615445
+      "required_token_positions_per_second": 9434.907894553542,
+      "required_pages_per_second_upper_estimate": 0.0037637258235812755
     },
     {
       "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": 18990.415470587195,
-      "required_pages_per_second_upper_estimate": 0.004393098399223092
+      "required_token_positions_per_second": 18869.815789107084,
+      "required_pages_per_second_upper_estimate": 0.004365199785394571
     },
     {
       "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": 28485.623205880795,
-      "required_pages_per_second_upper_estimate": 0.001734625000723478
+      "required_token_positions_per_second": 28304.723683660624,
+      "required_pages_per_second_upper_estimate": 0.0017236091689267119
     },
     {
       "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": 28485.623205880795,
-      "required_pages_per_second_upper_estimate": 0.00219290212764057
+      "required_token_positions_per_second": 28304.723683660624,
+      "required_pages_per_second_upper_estimate": 0.0021789759816581338
     },
     {
       "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": 28485.623205880795,
-      "required_pages_per_second_upper_estimate": 0.35369610496890613
+      "required_token_positions_per_second": 28304.723683660624,
+      "required_pages_per_second_upper_estimate": 0.35144993833468624
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.082133993078088,
-    "producer_p95_seconds": 0.20182093899893516,
-    "gpu_update_seconds": 0.9705948786927762,
-    "producer_batches_per_second": 12.175226876517021,
-    "gpu_batches_per_second": 1.0302959782219616,
-    "producer_headroom": 11.817212853270068,
+    "producer_batch_seconds": 0.6458460065937572,
+    "producer_p95_seconds": 0.26346012899921334,
+    "gpu_update_seconds": 0.976798088863177,
+    "producer_batches_per_second": 1.5483567131955787,
+    "gpu_batches_per_second": 1.0237530267527715,
+    "producer_headroom": 1.5124318783279116,
     "sustainable": true,
-    "buffer_seconds": 60.17688247895212,
+    "buffer_seconds": 60.56148150951697,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279776.53890904336,
-    "upload_bytes_per_second": 5675379.163774109,
+    "generation_bytes_per_second": 278158.1270721043,
+    "upload_bytes_per_second": 5877507.534260329,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.03353238538242,
+    "estimated_full_shards_per_hour": 59.68625887987468,
     "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": 0.9611707320000278,
-      "bytes": 976408,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 1.0036651360005635,
       "bytes": 7718969,
@@ -562,6 +556,12 @@
       "bytes": 5972956,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3616979350008478,
+      "bytes": 18065144,
+      "pages": 42,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 369100, "loss": 0.6122744921594858, "lr": 0.0003, "grad_norm": 0.1547183231806319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54025977342, "tokens_per_second": 147820.88459741397, "data_wait_s": 1.3742999726673588e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 369200, "loss": 0.5791829787194729, "lr": 0.0003, "grad_norm": 0.162201866261202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54040422815, "tokens_per_second": 147316.80578937265, "data_wait_s": 8.643000910524279e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 369300, "loss": 0.6052895896136761, "lr": 0.0003, "grad_norm": 0.17438512893661323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54054989261, "tokens_per_second": 146502.63530294, "data_wait_s": 1.5485999028896913e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 369400, "loss": 0.589466156437993, "lr": 0.0003, "grad_norm": 0.16202583969873532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54069685623, "tokens_per_second": 149994.51488694124, "data_wait_s": 1.4674000340164639e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c9d300be07d3ee675fe5478d83b16cd65fccfb39265b74cbf84347d9912dab55
-size 304605
+oid sha256:7a7851ad32f65152e0748b6dae9571c7b20c418459ad51b653d5701e91c8bd0b
+size 306723