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": 75874.97996176434,
-      "required_pages_per_second_upper_estimate": 0.0052492071679639055
+      "required_token_positions_per_second": 76058.0489074925,
+      "required_pages_per_second_upper_estimate": 0.005261872302407856
     },
     {
       "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": 9484.372495220543,
-      "required_pages_per_second_upper_estimate": 0.003970908707529417
+      "required_token_positions_per_second": 9507.256113436562,
+      "required_pages_per_second_upper_estimate": 0.0039804896006121766
     },
     {
       "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": 18968.744990441086,
-      "required_pages_per_second_upper_estimate": 0.003950689342063835
+      "required_token_positions_per_second": 19014.512226873125,
+      "required_pages_per_second_upper_estimate": 0.003960221450449439
     },
     {
       "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": 28453.11748566163,
-      "required_pages_per_second_upper_estimate": 0.001705720947870584
+      "required_token_positions_per_second": 28521.76834030969,
+      "required_pages_per_second_upper_estimate": 0.0017098364617829497
     },
     {
       "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": 28453.11748566163,
-      "required_pages_per_second_upper_estimate": 0.0022121047324728475
+      "required_token_positions_per_second": 28521.76834030969,
+      "required_pages_per_second_upper_estimate": 0.0022174420344585367
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28453.11748566163,
-      "required_pages_per_second_upper_estimate": 0.3649660405287468
+      "required_token_positions_per_second": 28521.76834030969,
+      "required_pages_per_second_upper_estimate": 0.36584661998062734
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.21790993567185524,
-    "producer_p95_seconds": 0.16944761099966854,
-    "gpu_update_seconds": 0.9717037162600073,
-    "producer_batches_per_second": 4.589051880157834,
-    "gpu_batches_per_second": 1.0291202794293124,
-    "producer_headroom": 4.459198766059341,
+    "producer_batch_seconds": 0.06730667486716158,
+    "producer_p95_seconds": 0.16763282399915624,
+    "gpu_update_seconds": 0.9693648609060892,
+    "producer_batches_per_second": 14.857367444961875,
+    "gpu_batches_per_second": 1.031603310919766,
+    "producer_headroom": 14.402209926716125,
     "sustainable": true,
-    "buffer_seconds": 60.245630408120455,
+    "buffer_seconds": 60.10062137617753,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 504758.88329942815,
-    "upload_bytes_per_second": 5316598.720717274,
+    "generation_bytes_per_second": 665024.8487394642,
+    "upload_bytes_per_second": 5285813.480805551,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.30950617062696,
+    "estimated_full_shards_per_hour": 71.34942577666256,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 6055828,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.397671879999507,
+      "bytes": 6213114,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 608100, "loss": 0.5482956767082214, "lr": 0.0001, "grad_norm": 0.13596161118764477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89007715322, "tokens_per_second": 150830.3868322628, "data_wait_s": 1.3473000308295013e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608200, "loss": 0.5756344199180603, "lr": 0.0001, "grad_norm": 0.1575766573149425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89022439208, "tokens_per_second": 153002.66190362268, "data_wait_s": 1.3965999642095994e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608300, "loss": 0.5493766069412231, "lr": 0.0001, "grad_norm": 0.17303658078893253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89037184969, "tokens_per_second": 152626.1485758194, "data_wait_s": 1.1683000593620818e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 608400, "loss": 0.5173752307891846, "lr": 0.0001, "grad_norm": 0.14122168409750036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89051901676, "tokens_per_second": 152737.989892279, "data_wait_s": 1.485699976910837e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:193f9f3489d1061cda5ddd7486788a533966b4ab599a9e252b2298040172b0f2
-size 83550
+oid sha256:57a0eda588633efca390b88941bfdeeef21d150ee6e5cb2fb82358a5eb9e5d4b
+size 85668