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": 76178.04433098817,
-      "required_pages_per_second_upper_estimate": 0.005270173864233028
+      "required_token_positions_per_second": 76053.50941191216,
+      "required_pages_per_second_upper_estimate": 0.005261558249570528
     },
     {
       "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": 9522.25554137352,
-      "required_pages_per_second_upper_estimate": 0.003986769547865708
+      "required_token_positions_per_second": 9506.68867648902,
+      "required_pages_per_second_upper_estimate": 0.003980252026611672
     },
     {
       "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": 19044.51108274704,
-      "required_pages_per_second_upper_estimate": 0.0039664694210049456
+      "required_token_positions_per_second": 19013.37735297804,
+      "required_pages_per_second_upper_estimate": 0.00395998508614573
     },
     {
       "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": 28566.766624120566,
-      "required_pages_per_second_upper_estimate": 0.0017125340401890107
+      "required_token_positions_per_second": 28520.066029467063,
+      "required_pages_per_second_upper_estimate": 0.0017097344108472125
     },
     {
       "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": 28566.766624120566,
-      "required_pages_per_second_upper_estimate": 0.0022209404530982996
+      "required_token_positions_per_second": 28520.066029467063,
+      "required_pages_per_second_upper_estimate": 0.002217309687278197
     },
     {
       "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": 28566.766624120566,
-      "required_pages_per_second_upper_estimate": 0.36642380964996046
+      "required_token_positions_per_second": 28520.066029467063,
+      "required_pages_per_second_upper_estimate": 0.3658247845649371
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5719319847501083,
-    "producer_p95_seconds": 0.31078248099947814,
-    "gpu_update_seconds": 0.9678379203285542,
-    "producer_batches_per_second": 1.74845965370677,
-    "gpu_batches_per_second": 1.03323085301362,
-    "producer_headroom": 1.6922255550219443,
+    "producer_batch_seconds": 0.11900169568758656,
+    "producer_p95_seconds": 0.3142748890022631,
+    "gpu_update_seconds": 0.9694227205306594,
+    "producer_batches_per_second": 8.403241602752331,
+    "gpu_batches_per_second": 1.0315417400704232,
+    "producer_headroom": 8.146293335816583,
     "sustainable": true,
-    "buffer_seconds": 60.00595106037036,
+    "buffer_seconds": 60.10420867290088,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360822.7539368064,
-    "upload_bytes_per_second": 5102321.647765909,
+    "generation_bytes_per_second": 360970.4538571368,
+    "upload_bytes_per_second": 5091956.917859195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.42416347101349,
+    "estimated_full_shards_per_hour": 77.45585643563821,
     "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": 1.7259603879974748,
-      "bytes": 12302933,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.298020391001046,
       "bytes": 6312632,
@@ -562,6 +556,12 @@
       "bytes": 5109029,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.8349382990018057,
+      "bytes": 11979278,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -282,3 +282,4 @@
 {"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}
 {"step": 953400, "loss": 0.5547279119491577, "lr": 3e-05, "grad_norm": 0.3430259935499054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139551962961, "tokens_per_second": 151408.32961208528, "data_wait_s": 8.521998097421601e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 953500, "loss": 0.568813681602478, "lr": 3e-05, "grad_norm": 0.35661222057761044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139566538487, "tokens_per_second": 151904.02677224466, "data_wait_s": 1.3524000678444281e-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 sha256:5e422379d3a27f3a7ee07b043293d7de603206b224e18478fc642a8aaaba4771
-size 605592
+oid sha256:6d9a9c182414b6c86e5fdf66605cc15551636d7f8f2253ce6e4933eba434bddb
+size 607710