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": 77215.26984690296,
-      "required_pages_per_second_upper_estimate": 0.0053419315321187886
+      "required_token_positions_per_second": 77055.58762642575,
+      "required_pages_per_second_upper_estimate": 0.005330884345592377
     },
     {
       "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": 9651.90873086287,
-      "required_pages_per_second_upper_estimate": 0.004041052630838426
+      "required_token_positions_per_second": 9631.94845330322,
+      "required_pages_per_second_upper_estimate": 0.004032695679442194
     },
     {
       "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": 19303.81746172574,
-      "required_pages_per_second_upper_estimate": 0.004020476101377134
+      "required_token_positions_per_second": 19263.89690660644,
+      "required_pages_per_second_upper_estimate": 0.004012161702521618
     },
     {
       "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": 28955.726192588612,
-      "required_pages_per_second_upper_estimate": 0.0017358515724117973
+      "required_token_positions_per_second": 28895.845359909657,
+      "required_pages_per_second_upper_estimate": 0.0017322618079254313
     },
     {
       "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": 28955.726192588612,
-      "required_pages_per_second_upper_estimate": 0.002251180348694357
+      "required_token_positions_per_second": 28895.845359909657,
+      "required_pages_per_second_upper_estimate": 0.0022465248773415156
     },
     {
       "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": 28955.726192588612,
-      "required_pages_per_second_upper_estimate": 0.3714129653620222
+      "required_token_positions_per_second": 28895.845359909657,
+      "required_pages_per_second_upper_estimate": 0.3706448783354454
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.28039768657809816,
-    "producer_p95_seconds": 0.21081521800078917,
-    "gpu_update_seconds": 0.954837043840975,
-    "producer_batches_per_second": 3.5663632328916295,
-    "gpu_batches_per_second": 1.0472991244425858,
-    "producer_headroom": 3.405295726557386,
+    "producer_batch_seconds": 0.09471287387506777,
+    "producer_p95_seconds": 0.22281435200056876,
+    "gpu_update_seconds": 0.9568157517329142,
+    "producer_batches_per_second": 10.558226765656618,
+    "gpu_batches_per_second": 1.045133295714325,
+    "producer_headroom": 10.102277679748312,
     "sustainable": true,
-    "buffer_seconds": 60.154733761981426,
+    "buffer_seconds": 60.2793923591736,
     "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": 361101.4480130434,
+    "generation_bytes_per_second": 361299.5731657501,
     "upload_bytes_per_second": 5457171.879664742,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.48396473210789,
+    "estimated_full_shards_per_hour": 77.52647777776123,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 569400, "loss": 0.5441634058952332, "lr": 0.0003, "grad_norm": 0.16145182076545128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83343526835, "tokens_per_second": 152547.85808011508, "data_wait_s": 1.3646000297740102e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 569500, "loss": 0.5509956479072571, "lr": 0.0003, "grad_norm": 0.13833991359536205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83358015784, "tokens_per_second": 150899.6004511357, "data_wait_s": 7.894999725976959e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 569600, "loss": 0.6090668439865112, "lr": 0.0003, "grad_norm": 0.19079312157672362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83372512482, "tokens_per_second": 151103.87350086385, "data_wait_s": 1.0824000128195621e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 569700, "loss": 0.550571858882904, "lr": 0.0003, "grad_norm": 0.15825983676964148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83387139040, "tokens_per_second": 131797.99033992435, "data_wait_s": 1.2139000318711624e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b5cd391c5d8b63a58b92d10fb3ac26fa7c7007c1b2947b614563de5fe0cc0af
-size 266481
+oid sha256:ecf18ca9ae479b3d423682da7619cefb113a61363a57d662593b09cf248816fb
+size 268599