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": 76484.64783261837,
-      "required_pages_per_second_upper_estimate": 0.005291385405893944
+      "required_token_positions_per_second": 76878.2009376511,
+      "required_pages_per_second_upper_estimate": 0.005318612322869122
     },
     {
       "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": 9560.580979077296,
-      "required_pages_per_second_upper_estimate": 0.004002815608306132
+      "required_token_positions_per_second": 9609.775117206387,
+      "required_pages_per_second_upper_estimate": 0.004023412166650361
     },
     {
       "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": 19121.161958154593,
-      "required_pages_per_second_upper_estimate": 0.00398243377693282
+      "required_token_positions_per_second": 19219.550234412774,
+      "required_pages_per_second_upper_estimate": 0.004002925460204069
     },
     {
       "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": 28681.742937231887,
-      "required_pages_per_second_upper_estimate": 0.0017194266946013695
+      "required_token_positions_per_second": 28829.325351619165,
+      "required_pages_per_second_upper_estimate": 0.001728274034998588
     },
     {
       "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": 28681.742937231887,
-      "required_pages_per_second_upper_estimate": 0.00222987935571535
+      "required_token_positions_per_second": 28829.325351619165,
+      "required_pages_per_second_upper_estimate": 0.002241353239287524
     },
     {
       "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": 28681.742937231887,
-      "required_pages_per_second_upper_estimate": 0.36789860234260574
+      "required_token_positions_per_second": 28829.325351619165,
+      "required_pages_per_second_upper_estimate": 0.3697916310927151
     }
   ],
   "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": 0.0660346452421976,
-    "producer_p95_seconds": 0.15174905200001376,
-    "gpu_update_seconds": 0.9639581548619912,
-    "producer_batches_per_second": 15.143565871100915,
-    "gpu_batches_per_second": 1.0373894291533525,
-    "producer_headroom": 14.597763815137462,
+    "producer_batch_seconds": 0.04851510257813629,
+    "producer_p95_seconds": 0.147574369999802,
+    "gpu_update_seconds": 0.959023482609772,
+    "producer_batches_per_second": 20.612138217979524,
+    "gpu_batches_per_second": 1.0427273347663182,
+    "producer_headroom": 19.767524577840703,
     "sustainable": true,
-    "buffer_seconds": 60.729363756305446,
+    "buffer_seconds": 60.418479404415635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317210.96635261126,
-    "upload_bytes_per_second": 5406677.8212594865,
+    "generation_bytes_per_second": 325747.7220814403,
+    "upload_bytes_per_second": 5278128.90318349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.06608908590081,
+    "estimated_full_shards_per_hour": 69.89787813980492,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 7362856,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.8067839350001123,
+      "bytes": 1162221,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -24,3 +24,4 @@
 {"step": 399500, "loss": 0.5737971067428589, "lr": 0.0003, "grad_norm": 0.14036649233757922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58475688757, "tokens_per_second": 152696.5091392368, "data_wait_s": 9.268999747291673e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 399600, "loss": 0.5494357347488403, "lr": 0.0003, "grad_norm": 0.16003251975744837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58490292297, "tokens_per_second": 151121.90553385427, "data_wait_s": 9.64700029726373e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 399700, "loss": 0.6012229323387146, "lr": 0.0003, "grad_norm": 0.19650151226153972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58504966712, "tokens_per_second": 152338.8856773339, "data_wait_s": 1.0697000107029453e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 399800, "loss": 0.5576823353767395, "lr": 0.0003, "grad_norm": 0.1638329301772677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58519792173, "tokens_per_second": 131438.76742203868, "data_wait_s": 1.1736000033124583e-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:f16672d32a652cd3e173ac8df392bc5a4a4d4460c76da7c4fb7e3b19bcc55444
-size 55233
+oid sha256:40ab1894bce8a8bda56ae355fc878a07369f5181bb9a1694475d9480dc521181
+size 57351