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": 76371.66331857373,
-      "required_pages_per_second_upper_estimate": 0.005283568875052145
+      "required_token_positions_per_second": 77086.57282946538,
+      "required_pages_per_second_upper_estimate": 0.005333027973834235
     },
     {
       "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": 9546.457914821716,
-      "required_pages_per_second_upper_estimate": 0.003996902576225439
+      "required_token_positions_per_second": 9635.821603683173,
+      "required_pages_per_second_upper_estimate": 0.004034317286625703
     },
     {
       "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": 19092.915829643433,
-      "required_pages_per_second_upper_estimate": 0.003976550853264446
+      "required_token_positions_per_second": 19271.643207366345,
+      "required_pages_per_second_upper_estimate": 0.004013775052686219
     },
     {
       "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": 28639.373744465145,
-      "required_pages_per_second_upper_estimate": 0.0017168867261890147
+      "required_token_positions_per_second": 28907.464811049518,
+      "required_pages_per_second_upper_estimate": 0.0017329583762793915
     },
     {
       "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": 28639.373744465145,
-      "required_pages_per_second_upper_estimate": 0.0022265853373401195
+      "required_token_positions_per_second": 28907.464811049518,
+      "required_pages_per_second_upper_estimate": 0.0022474282385590754
     },
     {
       "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": 28639.373744465145,
-      "required_pages_per_second_upper_estimate": 0.3673551358302888
+      "required_token_positions_per_second": 28907.464811049518,
+      "required_pages_per_second_upper_estimate": 0.3707939201786729
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2852437854765526,
+    "producer_batch_seconds": 0.2361399784842888,
     "producer_p95_seconds": 0.2791147669995553,
-    "gpu_update_seconds": 0.9653842380315059,
-    "producer_batches_per_second": 3.50577313482681,
-    "gpu_batches_per_second": 1.0358569786047869,
-    "producer_headroom": 3.384418126476104,
+    "gpu_update_seconds": 0.9564311564752609,
+    "producer_batches_per_second": 4.234776366198972,
+    "gpu_batches_per_second": 1.0455535594274277,
+    "producer_headroom": 4.050272057337786,
     "sustainable": true,
-    "buffer_seconds": 60.81920699598487,
+    "buffer_seconds": 60.255162857941436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357895.7613252709,
-    "upload_bytes_per_second": 4977110.438615836,
+    "generation_bytes_per_second": 358004.1787412563,
+    "upload_bytes_per_second": 4949615.387672751,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.79609899347874,
+    "estimated_full_shards_per_hour": 76.81936284712093,
     "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": 1.1762084339989087,
-      "bytes": 10342096,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.164775548999387,
       "bytes": 5765223,
@@ -562,6 +556,12 @@
       "bytes": 7640568,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9312373259999731,
+      "bytes": 6794347,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -231,3 +231,4 @@
 {"step": 420200, "loss": 0.5941126346588135, "lr": 0.0003, "grad_norm": 0.181814914446555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61506083731, "tokens_per_second": 150840.83369612007, "data_wait_s": 1.7055001080734655e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420300, "loss": 0.5611682534217834, "lr": 0.0003, "grad_norm": 0.14011688720046497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61520691441, "tokens_per_second": 151543.81931361408, "data_wait_s": 8.942999556893483e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 420400, "loss": 0.608125627040863, "lr": 0.0003, "grad_norm": 0.17101927890635338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61535371135, "tokens_per_second": 152043.47900095957, "data_wait_s": 1.3549997674999759e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 420500, "loss": 0.5629096031188965, "lr": 0.0003, "grad_norm": 0.15777687645875652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61550064028, "tokens_per_second": 152217.03841189836, "data_wait_s": 1.0747997293947265e-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 sha256:9e34d4fa10e1c3c592140dde15a0172ec80e0a63ab58564f6eb72eb27bf39c46
-size 497574
+oid sha256:a0a8c9b23a99a3b648178a580b2b9ad8300aa70ddd2b658bdfa638356947ac9d
+size 499692