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": 76748.75202943581,
-      "required_pages_per_second_upper_estimate": 0.0053096567470359395
+      "required_token_positions_per_second": 76984.1802698523,
+      "required_pages_per_second_upper_estimate": 0.005325944218976731
     },
     {
       "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": 9593.594003679476,
-      "required_pages_per_second_upper_estimate": 0.004016637472316718
+      "required_token_positions_per_second": 9623.022533731537,
+      "required_pages_per_second_upper_estimate": 0.0040289585833119266
     },
     {
       "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": 19187.188007358953,
-      "required_pages_per_second_upper_estimate": 0.003996185261758078
+      "required_token_positions_per_second": 19246.045067463074,
+      "required_pages_per_second_upper_estimate": 0.004008443635212713
     },
     {
       "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": 28780.78201103843,
-      "required_pages_per_second_upper_estimate": 0.0017253639358521622
+      "required_token_positions_per_second": 28869.067601194613,
+      "required_pages_per_second_upper_estimate": 0.0017306565221777346
     },
     {
       "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": 28780.78201103843,
-      "required_pages_per_second_upper_estimate": 0.002237579208076962
+      "required_token_positions_per_second": 28869.067601194613,
+      "required_pages_per_second_upper_estimate": 0.0022444430243843336
     },
     {
       "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": 28780.78201103843,
-      "required_pages_per_second_upper_estimate": 0.3691689692415237
+      "required_token_positions_per_second": 28869.067601194613,
+      "required_pages_per_second_upper_estimate": 0.37030140199836603
     }
   ],
   "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": 0.4350306173750482,
-    "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,
+    "producer_batch_seconds": 0.056205389679661266,
+    "producer_p95_seconds": 0.1989416260003054,
+    "gpu_update_seconds": 0.9577032546370121,
+    "producer_batches_per_second": 17.791888032436585,
+    "gpu_batches_per_second": 1.0441647714552449,
+    "producer_headroom": 17.039349074801823,
     "sustainable": true,
-    "buffer_seconds": 60.520384725194404,
+    "buffer_seconds": 60.335305042131765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 249311.50950395086,
-    "upload_bytes_per_second": 5456969.587774888,
+    "generation_bytes_per_second": 269322.44105548254,
+    "upload_bytes_per_second": 5430040.257701523,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.496446264637896,
+    "estimated_full_shards_per_hour": 57.790326344951225,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 8199304,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4808365860008053,
+      "bytes": 6083362,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -108,3 +108,4 @@
 {"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}
+{"step": 771100, "loss": 0.4970531463623047, "lr": 3e-05, "grad_norm": 0.23643469945190743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112866569411, "tokens_per_second": 150266.1260719092, "data_wait_s": 9.981000403058715e-06, "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 sha256:7a7f20fa62c9d2edefa9f4e42957b5504e820a94c1a0495e6478b4769cde0d6b
-size 234711
+oid sha256:ed1a15751ec9d3ca5fc13d6ef8842ce12387b1edc4db2b6b5e443420f278a71c
+size 236829