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": 76654.84132717534,
-      "required_pages_per_second_upper_estimate": 0.00530315978675071
+      "required_token_positions_per_second": 77217.95860239197,
+      "required_pages_per_second_upper_estimate": 0.005342117546462291
     },
     {
       "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": 9581.855165896917,
-      "required_pages_per_second_upper_estimate": 0.004011722666071968
+      "required_token_positions_per_second": 9652.244825298996,
+      "required_pages_per_second_upper_estimate": 0.004041193346560382
     },
     {
       "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": 19163.710331793834,
-      "required_pages_per_second_upper_estimate": 0.003991295481085805
+      "required_token_positions_per_second": 19304.489650597992,
+      "required_pages_per_second_upper_estimate": 0.004020616100592412
     },
     {
       "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": 28745.565497690754,
-      "required_pages_per_second_upper_estimate": 0.0017232527596494714
+      "required_token_positions_per_second": 28956.73447589699,
+      "required_pages_per_second_upper_estimate": 0.001735912017456575
     },
     {
       "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": 28745.565497690754,
-      "required_pages_per_second_upper_estimate": 0.002234841279065252
+      "required_token_positions_per_second": 28956.73447589699,
+      "required_pages_per_second_upper_estimate": 0.0022512587382866136
     },
     {
       "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": 28745.565497690754,
-      "required_pages_per_second_upper_estimate": 0.36871724962084573
+      "required_token_positions_per_second": 28956.73447589699,
+      "required_pages_per_second_upper_estimate": 0.37142589853769176
     }
   ],
   "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": 0.11992026080469032,
-    "producer_p95_seconds": 0.3475645110011101,
-    "gpu_update_seconds": 0.9618179194359935,
-    "producer_batches_per_second": 8.338874459493237,
-    "gpu_batches_per_second": 1.03969782616069,
-    "producer_headroom": 8.02047888306773,
+    "producer_batch_seconds": 0.13386163022670416,
+    "producer_p95_seconds": 0.3530330880021211,
+    "gpu_update_seconds": 0.954803796091498,
+    "producer_batches_per_second": 7.470400579362653,
+    "gpu_batches_per_second": 1.0473355930228947,
+    "producer_headroom": 7.132766831499587,
     "sustainable": true,
-    "buffer_seconds": 60.59452892446759,
+    "buffer_seconds": 60.152639153764376,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362636.82671076193,
+    "generation_bytes_per_second": 362000.8439708823,
     "upload_bytes_per_second": 4680960.055488144,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.81342125885146,
+    "estimated_full_shards_per_hour": 77.67695416779377,
     "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": [

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

@@ -383,3 +383,4 @@
 {"step": 642700, "loss": 0.506992518901825, "lr": 0.0001, "grad_norm": 0.17536772722896798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94074894927, "tokens_per_second": 154416.61329765926, "data_wait_s": 1.3270000636111945e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 642800, "loss": 0.5037686824798584, "lr": 0.0001, "grad_norm": 0.16527506314839135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94089365992, "tokens_per_second": 143680.36548489326, "data_wait_s": 1.077799970516935e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 642900, "loss": 0.5826767086982727, "lr": 0.0001, "grad_norm": 0.17394706545200048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94103865498, "tokens_per_second": 151333.37496133015, "data_wait_s": 1.0149000445380807e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 643000, "loss": 0.5572802424430847, "lr": 0.0001, "grad_norm": 0.1743637421779304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94118435350, "tokens_per_second": 150919.23249322307, "data_wait_s": 1.2418000551406294e-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 sha256:17d183c8f3dddd131ea4aa763472da01ef1452bf9cf5bdf2699d2b0c9ba31953
-size 821859
+oid sha256:880d140095e64552c34d888eed34bc94d571968c2281b392def55530858484a0
+size 823977