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": 76250.73798178947,
-      "required_pages_per_second_upper_estimate": 0.005275202979667971
+      "required_token_positions_per_second": 77261.42380576664,
+      "required_pages_per_second_upper_estimate": 0.005345124570084917
     },
     {
       "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": 9531.342247723684,
-      "required_pages_per_second_upper_estimate": 0.0039905739620625155
+      "required_token_positions_per_second": 9657.67797572083,
+      "required_pages_per_second_upper_estimate": 0.004043468093184921
     },
     {
       "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": 19062.684495447367,
-      "required_pages_per_second_upper_estimate": 0.003970254463605301
+      "required_token_positions_per_second": 19315.35595144166,
+      "required_pages_per_second_upper_estimate": 0.0040228792644945235
     },
     {
       "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": 28594.02674317105,
-      "required_pages_per_second_upper_estimate": 0.0017141682427022954
+      "required_token_positions_per_second": 28973.033927162487,
+      "required_pages_per_second_upper_estimate": 0.0017368891446721206
     },
     {
       "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": 28594.02674317105,
-      "required_pages_per_second_upper_estimate": 0.002223059807449883
+      "required_token_positions_per_second": 28973.033927162487,
+      "required_pages_per_second_upper_estimate": 0.0022525259489287967
     },
     {
       "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": 28594.02674317105,
-      "required_pages_per_second_upper_estimate": 0.36677347318750464
+      "required_token_positions_per_second": 28973.033927162487,
+      "required_pages_per_second_upper_estimate": 0.37163497039753834
     }
   ],
   "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": 0.4259989899367156,
-    "producer_p95_seconds": 0.3183238369965693,
-    "gpu_update_seconds": 0.9669152319235002,
-    "producer_batches_per_second": 2.3474234062117265,
-    "gpu_batches_per_second": 1.0342168237547398,
-    "producer_headroom": 2.2697594472398643,
+    "producer_batch_seconds": 0.1218054964294879,
+    "producer_p95_seconds": 0.30949456399685005,
+    "gpu_update_seconds": 0.9542666491022793,
+    "producer_batches_per_second": 8.20981014250774,
+    "gpu_batches_per_second": 1.0479251275738748,
+    "producer_headroom": 7.834348014456767,
     "sustainable": true,
-    "buffer_seconds": 60.91565961118051,
+    "buffer_seconds": 60.118798893443596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285505.5969631418,
-    "upload_bytes_per_second": 5597664.208611176,
+    "generation_bytes_per_second": 286416.33806753746,
+    "upload_bytes_per_second": 5540130.38895154,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.262854878146086,
+    "estimated_full_shards_per_hour": 61.458278718181546,
     "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": 1.162798567002028,
-      "bytes": 6350229,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0111840479985403,
       "bytes": 5239532,
@@ -562,6 +556,12 @@
       "bytes": 16775004,
       "pages": 44,
       "shards": 4
+    },
+    {
+      "seconds": 1.3269825160023174,
+      "bytes": 2735395,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 521700, "loss": 0.5954093933105469, "lr": 0.0003, "grad_norm": 0.20252088826755185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76359106129, "tokens_per_second": 153328.44297819925, "data_wait_s": 1.2363998393993825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521800, "loss": 0.5179608464241028, "lr": 0.0003, "grad_norm": 0.15108013226515238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76373767291, "tokens_per_second": 152419.8019010753, "data_wait_s": 8.96599522093311e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 521900, "loss": 0.580098032951355, "lr": 0.0003, "grad_norm": 0.17006697372490467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76388434426, "tokens_per_second": 153243.27918139452, "data_wait_s": 1.3111995940562338e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 522000, "loss": 0.6028584837913513, "lr": 0.0003, "grad_norm": 0.17199760238003273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76403168891, "tokens_per_second": 153939.4876438651, "data_wait_s": 1.3580000086221844e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51d01444a3bced46b9e8cb78c5669ddd27a95a32b144ef9cfe4e09389c699a3e
-size 787188
+oid sha256:943054ce911cdd3ffc3b961b550f4d606aa582d4564a4343656f9b1388b426b4
+size 789306