Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77022.5547403701,
-      "required_pages_per_second_upper_estimate": 0.005328599053888192
+      "required_token_positions_per_second": 76856.65636274971,
+      "required_pages_per_second_upper_estimate": 0.005317121819187143
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9627.819342546263,
-      "required_pages_per_second_upper_estimate": 0.004030966906993894
+      "required_token_positions_per_second": 9607.082045343714,
+      "required_pages_per_second_upper_estimate": 0.004022284633699195
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19255.638685092526,
-      "required_pages_per_second_upper_estimate": 0.004010441732764217
+      "required_token_positions_per_second": 19214.164090687427,
+      "required_pages_per_second_upper_estimate": 0.004001803668508242
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28883.45802763879,
-      "required_pages_per_second_upper_estimate": 0.0017315192062701541
+      "required_token_positions_per_second": 28821.246136031143,
+      "required_pages_per_second_upper_estimate": 0.001727789698360334
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28883.45802763879,
-      "required_pages_per_second_upper_estimate": 0.0022455618167435717
+      "required_token_positions_per_second": 28821.246136031143,
+      "required_pages_per_second_upper_estimate": 0.0022407251158122693
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28883.45802763879,
-      "required_pages_per_second_upper_estimate": 0.3704859869375558
+      "required_token_positions_per_second": 28821.246136031143,
+      "required_pages_per_second_upper_estimate": 0.3696879995899378
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3588128503671877,
-    "producer_p95_seconds": 0.12907270899995638,
-    "gpu_update_seconds": 0.9572261040746378,
-    "producer_batches_per_second": 2.7869681896193508,
-    "gpu_batches_per_second": 1.044685258522815,
-    "producer_headroom": 2.6677587023292775,
+    "producer_batch_seconds": 0.044568604203124806,
+    "producer_p95_seconds": 0.12861770300003172,
+    "gpu_update_seconds": 0.959292317532225,
+    "producer_batches_per_second": 22.43731922683564,
+    "gpu_batches_per_second": 1.04243511776733,
+    "producer_headroom": 21.523947960321514,
     "sustainable": true,
-    "buffer_seconds": 60.305244556702185,
+    "buffer_seconds": 60.435416004530175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 46,
+    "upload_pages_per_commit": 54,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 242871.18008123705,
-    "upload_bytes_per_second": 770242.1525414627,
+    "generation_bytes_per_second": 248306.76345571727,
+    "upload_bytes_per_second": 2357539.1999095096,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.11450149371942,
+    "estimated_full_shards_per_hour": 53.280851151977906,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 873319,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.4327509930000133,
+      "bytes": 9634876,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -2,3 +2,4 @@
 {"step": 397300, "loss": 0.5580515265464783, "lr": 0.0003, "grad_norm": 0.18260739819030208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58154046642, "tokens_per_second": 153952.47086143732, "data_wait_s": 1.124099992466654e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 397400, "loss": 0.5864430665969849, "lr": 0.0003, "grad_norm": 0.16819411572980625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58168497235, "tokens_per_second": 148897.31773298592, "data_wait_s": 1.0385999985373928e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 397500, "loss": 0.5782780051231384, "lr": 0.0003, "grad_norm": 0.14115527360265942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58183021172, "tokens_per_second": 149143.60326988477, "data_wait_s": 1.2393000019983447e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 397600, "loss": 0.5619044899940491, "lr": 0.0003, "grad_norm": 0.16381671188554098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58197534493, "tokens_per_second": 150408.54951693214, "data_wait_s": 1.1349000033078482e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:202e38262e34c7ffa2bf7bb7f0714346ea3f7cbb7842f8cdef714c4a1c29b8f9
-size 8637
+oid sha256:298555932b0e5c2ffa8f99b879ca545c48bbcb09f5b0cdbadf3f79eb6c222bc8
+size 10755