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": 76604.39951244507,
-      "required_pages_per_second_upper_estimate": 0.005299670104966535
+      "required_token_positions_per_second": 76819.46868266285,
+      "required_pages_per_second_upper_estimate": 0.005314549089191428
     },
     {
       "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": 9575.549939055634,
-      "required_pages_per_second_upper_estimate": 0.0040090827992616315
+      "required_token_positions_per_second": 9602.433585332856,
+      "required_pages_per_second_upper_estimate": 0.004020338420563532
     },
     {
       "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": 19151.09987811127,
-      "required_pages_per_second_upper_estimate": 0.003988669056143753
+      "required_token_positions_per_second": 19204.86717066571,
+      "required_pages_per_second_upper_estimate": 0.003999867365243987
     },
     {
       "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": 28726.649817166905,
-      "required_pages_per_second_upper_estimate": 0.0017221187934846384
+      "required_token_positions_per_second": 28807.300755998567,
+      "required_pages_per_second_upper_estimate": 0.0017269536941207433
     },
     {
       "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": 28726.649817166905,
-      "required_pages_per_second_upper_estimate": 0.0022333706681058082
+      "required_token_positions_per_second": 28807.300755998567,
+      "required_pages_per_second_upper_estimate": 0.0022396409238539805
     },
     {
       "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": 28726.649817166905,
-      "required_pages_per_second_upper_estimate": 0.3684746195811612
+      "required_token_positions_per_second": 28807.300755998567,
+      "required_pages_per_second_upper_estimate": 0.3695091232282624
     }
   ],
   "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": 0.14443124731261037,
-    "producer_p95_seconds": 0.30670689700127696,
-    "gpu_update_seconds": 0.962451249135139,
-    "producer_batches_per_second": 6.923709506126307,
-    "gpu_batches_per_second": 1.0390136652621131,
-    "producer_headroom": 6.6637328628201,
+    "producer_batch_seconds": 0.08416861108602802,
+    "producer_p95_seconds": 0.29658896300315973,
+    "gpu_update_seconds": 0.9597567031420962,
+    "producer_batches_per_second": 11.88091364579972,
+    "gpu_batches_per_second": 1.0419307275751797,
+    "producer_headroom": 11.402786511008681,
     "sustainable": true,
-    "buffer_seconds": 60.634428695513755,
+    "buffer_seconds": 60.464672297952056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408943.73773826286,
-    "upload_bytes_per_second": 5049816.024090298,
+    "generation_bytes_per_second": 409089.97733105376,
+    "upload_bytes_per_second": 5068248.870990279,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.74980639563479,
+    "estimated_full_shards_per_hour": 87.78118600796422,
     "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": 1.2550481799989939,
-      "bytes": 6009424,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9456898729986278,
       "bytes": 975770,
@@ -562,6 +556,12 @@
       "bytes": 1187408,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2242876330019499,
+      "bytes": 7412020,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 953000, "loss": 0.4624035656452179, "lr": 3e-05, "grad_norm": 0.35700770549621713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139493468077, "tokens_per_second": 153853.7115908976, "data_wait_s": 1.03069978649728e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 953100, "loss": 0.5442652702331543, "lr": 3e-05, "grad_norm": 0.333775333286914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139508205139, "tokens_per_second": 153124.25215727597, "data_wait_s": 1.4394998288480565e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 953200, "loss": 0.5914426445960999, "lr": 3e-05, "grad_norm": 0.35556754710838473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139522881195, "tokens_per_second": 152480.93473686764, "data_wait_s": 1.1143998563056812e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 953300, "loss": 0.4815470576286316, "lr": 3e-05, "grad_norm": 0.35038475854920687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139537409847, "tokens_per_second": 138880.933205589, "data_wait_s": 1.095400148187764e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14bec3e8b624e810e0686595d7523a80849f74d8f0a74af83a9bdb8a58199804
-size 601356
+oid sha256:1af51bba82a8b998233c9b4bdf502c3e0a8c8355a6151850c23786bf2562ace8
+size 603474