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": 77245.30308392389,
-      "required_pages_per_second_upper_estimate": 0.005344009301142608
+      "required_token_positions_per_second": 77087.92080953546,
+      "required_pages_per_second_upper_estimate": 0.005333121230223227
     },
     {
       "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": 9655.662885490487,
-      "required_pages_per_second_upper_estimate": 0.004042624416985345
+      "required_token_positions_per_second": 9635.990101191932,
+      "required_pages_per_second_upper_estimate": 0.004034387833013992
     },
     {
       "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": 19311.325770980973,
-      "required_pages_per_second_upper_estimate": 0.004022039884187569
+      "required_token_positions_per_second": 19271.980202383864,
+      "required_pages_per_second_upper_estimate": 0.004013845239861211
     },
     {
       "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": 28966.988656471458,
-      "required_pages_per_second_upper_estimate": 0.0017365267399247874
+      "required_token_positions_per_second": 28907.970303575792,
+      "required_pages_per_second_upper_estimate": 0.001732988679784499
     },
     {
       "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": 28966.988656471458,
-      "required_pages_per_second_upper_estimate": 0.0022520559557229047
+      "required_token_positions_per_second": 28907.970303575792,
+      "required_pages_per_second_upper_estimate": 0.0022474675383795666
     },
     {
       "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": 28966.988656471458,
-      "required_pages_per_second_upper_estimate": 0.3715574281560198
+      "required_token_positions_per_second": 28907.970303575792,
+      "required_pages_per_second_upper_estimate": 0.37080040409404436
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09375373141413945,
-    "producer_p95_seconds": 0.1944217110012687,
-    "gpu_update_seconds": 0.9544657999451115,
-    "producer_batches_per_second": 10.666242131555151,
-    "gpu_batches_per_second": 1.0477064762902002,
-    "producer_headroom": 10.18056332850304,
+    "producer_batch_seconds": 0.2957698886718845,
+    "producer_p95_seconds": 0.1900108800000453,
+    "gpu_update_seconds": 0.9564144320633974,
+    "producer_batches_per_second": 3.381006783653223,
+    "gpu_batches_per_second": 1.0455718425772496,
+    "producer_headroom": 3.2336436827901913,
     "sustainable": true,
-    "buffer_seconds": 60.131345396542024,
+    "buffer_seconds": 60.254109219994035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329296.466331183,
+    "generation_bytes_per_second": 341082.03567337437,
     "upload_bytes_per_second": 4011819.1440317766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.65935604526155,
+    "estimated_full_shards_per_hour": 73.18826487208294,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 933600, "loss": 0.5904932618141174, "lr": 3e-05, "grad_norm": 0.3322589523926682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136652547151, "tokens_per_second": 152252.65762514182, "data_wait_s": 1.0293999366695061e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 933700, "loss": 0.5050845146179199, "lr": 3e-05, "grad_norm": 0.34105192537751106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136667142011, "tokens_per_second": 152649.665056586, "data_wait_s": 7.546999768237583e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 933800, "loss": 0.5353503227233887, "lr": 3e-05, "grad_norm": 0.3380934524447216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136681821891, "tokens_per_second": 152289.93964229143, "data_wait_s": 7.972001185407862e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 933900, "loss": 0.5196212530136108, "lr": 3e-05, "grad_norm": 0.33102221519367003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136696449396, "tokens_per_second": 153363.11647315358, "data_wait_s": 1.1234000339754857e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:899acc3b2f2c7748408fdb25f06dbf8053563da9e72da2076133d200c7cd5912
-size 187332
+oid sha256:bf71d2e18bcac9464532fcbf453844273bb4ee16465116c587c448edc785a5ef
+size 189450