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": 76853.96636159335,
-      "required_pages_per_second_upper_estimate": 0.005316935718665484
+      "required_token_positions_per_second": 76851.94686285881,
+      "required_pages_per_second_upper_estimate": 0.005316796005057145
     },
     {
       "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": 9606.74579519917,
-      "required_pages_per_second_upper_estimate": 0.004022143852785376
+      "required_token_positions_per_second": 9606.493357857351,
+      "required_pages_per_second_upper_estimate": 0.0040220381625418475
     },
     {
       "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": 19213.49159039834,
-      "required_pages_per_second_upper_estimate": 0.0040016636044330495
+      "required_token_positions_per_second": 19212.986715714702,
+      "required_pages_per_second_upper_estimate": 0.004001558452350889
     },
     {
       "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": 28820.23738559751,
-      "required_pages_per_second_upper_estimate": 0.0017277292253121102
+      "required_token_positions_per_second": 28819.480073572053,
+      "required_pages_per_second_upper_estimate": 0.0017276838256125326
     },
     {
       "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": 28820.23738559751,
-      "required_pages_per_second_upper_estimate": 0.0022406466899030785
+      "required_token_positions_per_second": 28819.480073572053,
+      "required_pages_per_second_upper_estimate": 0.0022405878122241625
     },
     {
       "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": 28820.23738559751,
-      "required_pages_per_second_upper_estimate": 0.369675060422487
+      "required_token_positions_per_second": 28819.480073572053,
+      "required_pages_per_second_upper_estimate": 0.36966534643696275
     }
   ],
   "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": 0.10431916198432845,
-    "producer_p95_seconds": 0.17889686000035,
-    "gpu_update_seconds": 0.9593258941654895,
-    "producer_batches_per_second": 9.585966575826472,
-    "gpu_batches_per_second": 1.0423986322915766,
-    "producer_headroom": 9.196065956795225,
+    "producer_batch_seconds": 0.057461221882782354,
+    "producer_p95_seconds": 0.17391686100017978,
+    "gpu_update_seconds": 0.9593511031225606,
+    "producer_batches_per_second": 17.40304099414982,
+    "gpu_batches_per_second": 1.0423712410869521,
+    "producer_headroom": 16.69562657542477,
     "sustainable": true,
-    "buffer_seconds": 60.43753133242584,
+    "buffer_seconds": 60.43911949672132,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284212.28648378915,
-    "upload_bytes_per_second": 5288285.566545509,
+    "generation_bytes_per_second": 300703.33163989556,
+    "upload_bytes_per_second": 5291223.900787543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.985340556004104,
+    "estimated_full_shards_per_hour": 64.52393495462084,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 6772198,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1199771049996343,
+      "bytes": 6074153,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 891100, "loss": 0.5342656970024109, "lr": 3e-05, "grad_norm": 0.3095160137590757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130432070713, "tokens_per_second": 152782.13955704786, "data_wait_s": 1.8817000636772718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891200, "loss": 0.4989605247974396, "lr": 3e-05, "grad_norm": 0.31334013940086874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130446723013, "tokens_per_second": 153155.12638318937, "data_wait_s": 1.2485999832279049e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891300, "loss": 0.5515876412391663, "lr": 3e-05, "grad_norm": 0.32950961351871944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130461425008, "tokens_per_second": 153256.2870949149, "data_wait_s": 1.0466000276210252e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 891400, "loss": 0.5025626420974731, "lr": 3e-05, "grad_norm": 0.3141751969306708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130476014189, "tokens_per_second": 152513.9970211436, "data_wait_s": 9.995999789680354e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ccd833006c6f70486e2d38041ea28f910181897d8bb3e3ace34738424361e06
-size 134382
+oid sha256:831f6e587d9c98fafc3450c43fdd39d0f06f656ab6e40f78e5ef0c2df901d9ca
+size 136500