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": 77385.78071356808,
-      "required_pages_per_second_upper_estimate": 0.0053537278695144006
+      "required_token_positions_per_second": 76748.75202943581,
+      "required_pages_per_second_upper_estimate": 0.0053096567470359395
     },
     {
       "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": 9673.22258919601,
-      "required_pages_per_second_upper_estimate": 0.0040499762982385374
+      "required_token_positions_per_second": 9593.594003679476,
+      "required_pages_per_second_upper_estimate": 0.004016637472316718
     },
     {
       "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": 19346.44517839202,
-      "required_pages_per_second_upper_estimate": 0.004029354330590235
+      "required_token_positions_per_second": 19187.188007358953,
+      "required_pages_per_second_upper_estimate": 0.003996185261758078
     },
     {
       "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": 29019.667767588027,
-      "required_pages_per_second_upper_estimate": 0.0017396847721998804
+      "required_token_positions_per_second": 28780.78201103843,
+      "required_pages_per_second_upper_estimate": 0.0017253639358521622
     },
     {
       "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": 29019.667767588027,
-      "required_pages_per_second_upper_estimate": 0.002256151524901296
+      "required_token_positions_per_second": 28780.78201103843,
+      "required_pages_per_second_upper_estimate": 0.002237579208076962
     },
     {
       "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": 29019.667767588027,
-      "required_pages_per_second_upper_estimate": 0.3722331392310005
+      "required_token_positions_per_second": 28780.78201103843,
+      "required_pages_per_second_upper_estimate": 0.3691689692415237
     }
   ],
   "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": 0.0893907867422854,
-    "producer_p95_seconds": 0.19242108300022664,
-    "gpu_update_seconds": 0.9527331677752686,
-    "producer_batches_per_second": 11.186835203531777,
-    "gpu_batches_per_second": 1.0496118260846365,
-    "producer_headroom": 10.658068940840723,
+    "producer_batch_seconds": 0.4350306173750482,
+    "producer_p95_seconds": 0.22971766899900103,
+    "gpu_update_seconds": 0.9606410273840381,
+    "producer_batches_per_second": 2.2986887820309,
+    "gpu_batches_per_second": 1.0409715715798042,
+    "producer_headroom": 2.208214753206327,
     "sustainable": true,
-    "buffer_seconds": 60.02218956984192,
+    "buffer_seconds": 60.520384725194404,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263742.0735565503,
-    "upload_bytes_per_second": 5457112.558486102,
+    "generation_bytes_per_second": 249311.50950395086,
+    "upload_bytes_per_second": 5456969.587774888,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.592909383987255,
+    "estimated_full_shards_per_hour": 53.496446264637896,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 10313766,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5044031500001438,
+      "bytes": 8199304,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 770700, "loss": 0.6074251532554626, "lr": 3e-05, "grad_norm": 0.2844986677756334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112808199028, "tokens_per_second": 153507.88402762378, "data_wait_s": 1.2989999959245324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770800, "loss": 0.568845272064209, "lr": 3e-05, "grad_norm": 0.24240980720523625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112822968721, "tokens_per_second": 154027.62018610426, "data_wait_s": 8.985000022221357e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 770900, "loss": 0.5253165364265442, "lr": 3e-05, "grad_norm": 0.24298080124792695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112837598704, "tokens_per_second": 151614.51904652428, "data_wait_s": 1.4668999938294291e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 771000, "loss": 0.492196261882782, "lr": 3e-05, "grad_norm": 0.23602603074079667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112852090321, "tokens_per_second": 150908.68533219758, "data_wait_s": 1.5677000192226842e-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 sha256:8793173b05f1a2580311d3422fae13db32fa9088780f18492eafd867cf6f2e3e
-size 232593
+oid sha256:7a7f20fa62c9d2edefa9f4e42957b5504e820a94c1a0495e6478b4769cde0d6b
+size 234711