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": 77082.16792079688,
-      "required_pages_per_second_upper_estimate": 0.0053327232320317525
+      "required_token_positions_per_second": 76346.72945978206,
+      "required_pages_per_second_upper_estimate": 0.005281843892846414
     },
     {
       "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": 9635.27099009961,
-      "required_pages_per_second_upper_estimate": 0.004034086756216385
+      "required_token_positions_per_second": 9543.341182472757,
+      "required_pages_per_second_upper_estimate": 0.003995597665475702
     },
     {
       "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": 19270.54198019922,
-      "required_pages_per_second_upper_estimate": 0.00401354569610862
+      "required_token_positions_per_second": 19086.682364945515,
+      "required_pages_per_second_upper_estimate": 0.003975252586955388
     },
     {
       "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": 28905.812970298834,
-      "required_pages_per_second_upper_estimate": 0.0017328593509486026
+      "required_token_positions_per_second": 28630.02354741827,
+      "required_pages_per_second_upper_estimate": 0.0017163261961529788
     },
     {
       "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": 28905.812970298834,
-      "required_pages_per_second_upper_estimate": 0.0022472998152063893
+      "required_token_positions_per_second": 28630.02354741827,
+      "required_pages_per_second_upper_estimate": 0.0022258584006468957
     },
     {
       "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": 28905.812970298834,
-      "required_pages_per_second_upper_estimate": 0.37077273213913153
+      "required_token_positions_per_second": 28630.02354741827,
+      "required_pages_per_second_upper_estimate": 0.3672352015420309
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09559652269535945,
-    "producer_p95_seconds": 0.2528797359991586,
-    "gpu_update_seconds": 0.9564858123315454,
-    "producer_batches_per_second": 10.46063153559186,
-    "gpu_batches_per_second": 1.0454938140299057,
-    "producer_headroom": 10.00544565182156,
+    "producer_batch_seconds": 0.48809071673414905,
+    "producer_p95_seconds": 0.2559190940010012,
+    "gpu_update_seconds": 0.9656995200932406,
+    "producer_batches_per_second": 2.0487994664005775,
+    "gpu_batches_per_second": 1.03551879150095,
+    "producer_headroom": 1.978524661470325,
     "sustainable": true,
-    "buffer_seconds": 60.25860617688736,
+    "buffer_seconds": 60.83906976587416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290388.26851347263,
-    "upload_bytes_per_second": 4871312.236857659,
+    "generation_bytes_per_second": 288017.00397111656,
+    "upload_bytes_per_second": 4937239.79899042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.310562530070634,
+    "estimated_full_shards_per_hour": 61.80174435949442,
     "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": 1.0089250809996884,
-      "bytes": 2580280,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1785594149987446,
       "bytes": 10228291,
@@ -562,6 +556,12 @@
       "bytes": 6170248,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.596614274996682,
+      "bytes": 11247081,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 781200, "loss": 0.5123441815376282, "lr": 3e-05, "grad_norm": 0.24898188702796165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114345432347, "tokens_per_second": 150619.8920894047, "data_wait_s": 1.2929001968586817e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 781300, "loss": 0.5730502009391785, "lr": 3e-05, "grad_norm": 0.25821237201401737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114360065856, "tokens_per_second": 151143.15535135104, "data_wait_s": 9.031999070430174e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 781400, "loss": 0.6186596155166626, "lr": 3e-05, "grad_norm": 0.3043311171630653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114374771278, "tokens_per_second": 146093.55331686113, "data_wait_s": 1.126800270867534e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 781500, "loss": 0.5546749234199524, "lr": 3e-05, "grad_norm": 0.2590823139928074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114389535494, "tokens_per_second": 152849.77736903654, "data_wait_s": 9.603001672076061e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f16a52350e2ef1bd4eec6ccb49b03f985c14f9933ea8d5f1b601081329ba3f1
-size 456549
+oid sha256:9b5851576d0a97a0338cdfb7603a41ac249463b508e62d46f704d6de6e2b9664
+size 458667