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": 76528.66632985527,
-      "required_pages_per_second_upper_estimate": 0.005294430707669779
+      "required_token_positions_per_second": 76971.22784285364,
+      "required_pages_per_second_upper_estimate": 0.0053250481400231925
     },
     {
       "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": 9566.08329123191,
-      "required_pages_per_second_upper_estimate": 0.004005119311503925
+      "required_token_positions_per_second": 9621.403480356705,
+      "required_pages_per_second_upper_estimate": 0.004028280719473564
     },
     {
       "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": 19132.16658246382,
-      "required_pages_per_second_upper_estimate": 0.003984725749964972
+      "required_token_positions_per_second": 19242.80696071341,
+      "required_pages_per_second_upper_estimate": 0.004007769222971375
     },
     {
       "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": 28698.24987369573,
-      "required_pages_per_second_upper_estimate": 0.0017204162602377443
+      "required_token_positions_per_second": 28864.210441070114,
+      "required_pages_per_second_upper_estimate": 0.001730365342844725
     },
     {
       "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": 28698.24987369573,
-      "required_pages_per_second_upper_estimate": 0.002231162697419074
+      "required_token_positions_per_second": 28864.210441070114,
+      "required_pages_per_second_upper_estimate": 0.0022440654015490435
     },
     {
       "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": 28698.24987369573,
-      "required_pages_per_second_upper_estimate": 0.36811033559979645
+      "required_token_positions_per_second": 28864.210441070114,
+      "required_pages_per_second_upper_estimate": 0.37023909956350115
     }
   ],
   "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": 0.06603346498448559,
-    "producer_p95_seconds": 0.22812866100139217,
-    "gpu_update_seconds": 0.9634036961027937,
-    "producer_batches_per_second": 15.14383654159217,
-    "gpu_batches_per_second": 1.0379864682326292,
-    "producer_headroom": 14.589628097346447,
+    "producer_batch_seconds": 0.12648628899208347,
+    "producer_p95_seconds": 0.2599107930000173,
+    "gpu_update_seconds": 0.9578644133172061,
+    "producer_batches_per_second": 7.905995250304071,
+    "gpu_batches_per_second": 1.0439890929206495,
+    "producer_headroom": 7.572871502121126,
     "sustainable": true,
-    "buffer_seconds": 60.694432854476005,
+    "buffer_seconds": 60.345458038983985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343841.1233995657,
-    "upload_bytes_per_second": 5099116.791732661,
+    "generation_bytes_per_second": 341882.36627621274,
+    "upload_bytes_per_second": 5075369.83616498,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.78030086984852,
+    "estimated_full_shards_per_hour": 73.35999718870913,
     "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": 1.060693833998812,
-      "bytes": 7602919,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 1.202748789000907,
       "bytes": 6204281,
@@ -562,6 +556,12 @@
       "bytes": 10490550,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.171707494999282,
+      "bytes": 6154958,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -180,3 +180,4 @@
 {"step": 778000, "loss": 0.5105672478675842, "lr": 3e-05, "grad_norm": 0.2628503965985092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113877148301, "tokens_per_second": 151913.99886996634, "data_wait_s": 7.556001946795732e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 778100, "loss": 0.511470377445221, "lr": 3e-05, "grad_norm": 0.2513624045672413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113891846413, "tokens_per_second": 152039.9511622395, "data_wait_s": 1.0210998880211264e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778200, "loss": 0.5231640338897705, "lr": 3e-05, "grad_norm": 0.24825772797739806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113906537906, "tokens_per_second": 152285.53860219131, "data_wait_s": 1.3607001164928079e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 778300, "loss": 0.4791816771030426, "lr": 3e-05, "grad_norm": 0.24315523523943586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113921334291, "tokens_per_second": 153813.61661321798, "data_wait_s": 1.4381999790202826e-05, "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 sha256:985bee127869bf916734b03a77afc056d9c86d7c70866fa5b31ca041de6022dd
-size 387990
+oid sha256:88c86ef2d1a7f78481e663fe14dc29559ce0b61f7ae5004cbcb9ca765819d020
+size 390108