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": 77331.3246144179,
-      "required_pages_per_second_upper_estimate": 0.005349960470219633
+      "required_token_positions_per_second": 75837.76193142973,
+      "required_pages_per_second_upper_estimate": 0.0052466323382676285
     },
     {
       "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": 9666.415576802237,
-      "required_pages_per_second_upper_estimate": 0.004047126344295848
+      "required_token_positions_per_second": 9479.720241428717,
+      "required_pages_per_second_upper_estimate": 0.003968960906016886
     },
     {
       "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": 19332.831153604475,
-      "required_pages_per_second_upper_estimate": 0.004026518888252967
+      "required_token_positions_per_second": 18959.440482857433,
+      "required_pages_per_second_upper_estimate": 0.003948751458510526
     },
     {
       "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": 28999.24673040671,
-      "required_pages_per_second_upper_estimate": 0.0017384605622019788
+      "required_token_positions_per_second": 28439.16072428615,
+      "required_pages_per_second_upper_estimate": 0.0017048842613368647
     },
     {
       "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": 28999.24673040671,
-      "required_pages_per_second_upper_estimate": 0.0022545638790830986
+      "required_token_positions_per_second": 28439.16072428615,
+      "required_pages_per_second_upper_estimate": 0.0022110196556652095
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28999.24673040671,
-      "required_pages_per_second_upper_estimate": 0.37197120009243995
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28439.16072428615,
+      "required_pages_per_second_upper_estimate": 0.364787018179425
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.042553780859377,
-    "producer_p95_seconds": 0.13049337699999342,
-    "gpu_update_seconds": 0.9534040748379204,
-    "producer_batches_per_second": 23.499674524916944,
-    "gpu_batches_per_second": 1.0488732179689928,
-    "producer_headroom": 22.40468544942069,
+    "producer_batch_seconds": 0.0993961561875043,
+    "producer_p95_seconds": 0.13271451599985085,
+    "gpu_update_seconds": 0.9721805881700818,
+    "producer_batches_per_second": 10.060751223755231,
+    "gpu_batches_per_second": 1.0286154775855814,
+    "producer_headroom": 9.78086704214323,
     "sustainable": true,
-    "buffer_seconds": 60.06445671478899,
+    "buffer_seconds": 60.27519646654507,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 65,
+    "upload_pages_per_commit": 64,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312285.0804309098,
-    "upload_bytes_per_second": 3904223.0874083876,
+    "generation_bytes_per_second": 277330.01476000005,
+    "upload_bytes_per_second": 4306545.447321214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.0091086358592,
+    "estimated_full_shards_per_hour": 59.50856525516511,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 825792,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5516937659999712,
+      "bytes": 10821331,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 605100, "loss": 0.5622690320014954, "lr": 0.0001, "grad_norm": 0.14328663181039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88568593606, "tokens_per_second": 139609.19308701303, "data_wait_s": 1.6875000028448994e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605200, "loss": 0.5281063318252563, "lr": 0.0001, "grad_norm": 0.1335078791685974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88583296602, "tokens_per_second": 152306.21731186987, "data_wait_s": 1.2056000059601502e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605300, "loss": 0.5635167956352234, "lr": 0.0001, "grad_norm": 0.13613936133076468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88597999350, "tokens_per_second": 150328.2228645332, "data_wait_s": 7.193999863375211e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 605400, "loss": 0.5422095060348511, "lr": 0.0001, "grad_norm": 0.15087026951664284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88612490448, "tokens_per_second": 150471.31610519273, "data_wait_s": 1.3342000102056772e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e82e1645129f94656570f3e92b8b158b48371ae8e40058e24e29deeaed6d5682
-size 20010
+oid sha256:08e54eb1b20770a17c2ff900c11129127ac793990c165232658cef5e85807cf3
+size 22128