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": 75877.76814803886,
-      "required_pages_per_second_upper_estimate": 0.006576892632626824
+      "required_token_positions_per_second": 75799.32321637654,
+      "required_pages_per_second_upper_estimate": 0.006570093224767196
     },
     {
       "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": 9484.721018504857,
-      "required_pages_per_second_upper_estimate": 0.0037835970234980283
+      "required_token_positions_per_second": 9474.915402047067,
+      "required_pages_per_second_upper_estimate": 0.0037796854164859847
     },
     {
       "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": 18969.442037009714,
-      "required_pages_per_second_upper_estimate": 0.0043882465644847655
+      "required_token_positions_per_second": 18949.830804094134,
+      "required_pages_per_second_upper_estimate": 0.004383709850895657
     },
     {
       "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": 28454.16305551457,
-      "required_pages_per_second_upper_estimate": 0.0017327092426248276
+      "required_token_positions_per_second": 28424.7462061412,
+      "required_pages_per_second_upper_estimate": 0.0017309179108362701
     },
     {
       "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": 28454.16305551457,
-      "required_pages_per_second_upper_estimate": 0.002190480238178108
+      "required_token_positions_per_second": 28424.7462061412,
+      "required_pages_per_second_upper_estimate": 0.0021882156476822952
     },
     {
       "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": 28454.16305551457,
-      "required_pages_per_second_upper_estimate": 0.3533054751917078
+      "required_token_positions_per_second": 28424.7462061412,
+      "required_pages_per_second_upper_estimate": 0.3529402163743522
     }
   ],
   "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": 0.0803273764528285,
-    "producer_p95_seconds": 0.3147048190003261,
-    "gpu_update_seconds": 0.9716680102682432,
-    "producer_batches_per_second": 12.449055902967784,
-    "gpu_batches_per_second": 1.029158096625961,
-    "producer_headroom": 12.096349378954834,
+    "producer_batch_seconds": 0.09329332912483324,
+    "producer_p95_seconds": 0.2895079039990378,
+    "gpu_update_seconds": 0.9726735922105302,
+    "producer_batches_per_second": 10.718880003327222,
+    "gpu_batches_per_second": 1.028094119145732,
+    "producer_headroom": 10.425971517309907,
     "sustainable": true,
-    "buffer_seconds": 60.24341663663108,
+    "buffer_seconds": 60.30576271705287,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 169,
+    "upload_pages_per_commit": 174,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 742855.254101377,
-    "upload_bytes_per_second": 4890251.469037776,
+    "generation_bytes_per_second": 751784.2829597088,
+    "upload_bytes_per_second": 4857774.833885393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.6997223724412,
+    "estimated_full_shards_per_hour": 80.65770324036335,
     "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": 1.048674845998903,
-      "bytes": 7634002,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2676819320004142,
       "bytes": 6003365,
@@ -562,6 +556,12 @@
       "bytes": 2508707,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2800720170016575,
+      "bytes": 6017394,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -267,3 +267,4 @@
 {"step": 381700, "loss": 0.613114608451724, "lr": 0.0003, "grad_norm": 0.15971321352573042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55869369087, "tokens_per_second": 149673.83336199875, "data_wait_s": 1.0146999557036906e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381800, "loss": 0.6879843585193157, "lr": 0.0003, "grad_norm": 0.16391770986705295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55884028395, "tokens_per_second": 149425.7625419428, "data_wait_s": 1.1308999091852456e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381900, "loss": 0.6026112791150808, "lr": 0.0003, "grad_norm": 0.1552691653762093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55898769496, "tokens_per_second": 150092.91202039964, "data_wait_s": 1.5804002032382414e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 382000, "loss": 0.6758288592100143, "lr": 0.0003, "grad_norm": 0.1708545792673141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55913492207, "tokens_per_second": 148949.50972857204, "data_wait_s": 1.4430999726755545e-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:eb36b8efae3f7d347bf62bce6fbb5bd103f6a1a871ee0a06f432b220f81f4baf
-size 573822
+oid sha256:df9169fb765e126f73687c7aac92f6c0b6ed874d96d38fff44553ca38f0d31ec
+size 575940