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": 77167.94538282955,
-      "required_pages_per_second_upper_estimate": 0.005338657515886305
+      "required_token_positions_per_second": 76495.66838071843,
+      "required_pages_per_second_upper_estimate": 0.005292147832982701
     },
     {
       "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": 9645.993172853694,
-      "required_pages_per_second_upper_estimate": 0.004038575910230882
+      "required_token_positions_per_second": 9561.958547589804,
+      "required_pages_per_second_upper_estimate": 0.004003392367475417
     },
     {
       "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": 19291.986345707388,
-      "required_pages_per_second_upper_estimate": 0.004018011991918023
+      "required_token_positions_per_second": 19123.917095179608,
+      "required_pages_per_second_upper_estimate": 0.003983007599317281
     },
     {
       "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": 28937.979518561082,
-      "required_pages_per_second_upper_estimate": 0.0017347876864014491
+      "required_token_positions_per_second": 28685.87564276941,
+      "required_pages_per_second_upper_estimate": 0.001719674443988068
     },
     {
       "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": 28937.979518561082,
-      "required_pages_per_second_upper_estimate": 0.0022498006228480858
+      "required_token_positions_per_second": 28685.87564276941,
+      "required_pages_per_second_upper_estimate": 0.0022302006553930422
     },
     {
       "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": 28937.979518561082,
-      "required_pages_per_second_upper_estimate": 0.371185330082491
+      "required_token_positions_per_second": 28685.87564276941,
+      "required_pages_per_second_upper_estimate": 0.3679516122518876
     }
   ],
   "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": 0.07155008578132538,
-    "producer_p95_seconds": 0.2474813480002922,
-    "gpu_update_seconds": 0.9554226127731662,
-    "producer_batches_per_second": 13.97622363523428,
-    "gpu_batches_per_second": 1.0466572453183263,
-    "producer_headroom": 13.353200102277615,
+    "producer_batch_seconds": 0.12018325653119177,
+    "producer_p95_seconds": 0.24227042599886772,
+    "gpu_update_seconds": 0.9638192797147185,
+    "producer_batches_per_second": 8.320626590281025,
+    "gpu_batches_per_second": 1.0375389049034074,
+    "producer_headroom": 8.019580327019792,
     "sustainable": true,
-    "buffer_seconds": 60.19162460470947,
+    "buffer_seconds": 60.72061462202726,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "upload_pages_per_commit": 171,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 738246.5485056872,
-    "upload_bytes_per_second": 5028977.011009334,
+    "generation_bytes_per_second": 722265.2810049609,
+    "upload_bytes_per_second": 4936579.641696096,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.2052619046114,
+    "estimated_full_shards_per_hour": 77.4906579142171,
     "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": 1.369053440999778,
-      "bytes": 11976832,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.3680221860013262,
       "bytes": 5930978,
@@ -562,6 +556,12 @@
       "bytes": 6186205,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0562214180026785,
+      "bytes": 2441509,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 905600, "loss": 0.5333682298660278, "lr": 3e-05, "grad_norm": 0.34868177279533796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132554316032, "tokens_per_second": 150367.3079037568, "data_wait_s": 1.9656003132695332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905700, "loss": 0.5601820945739746, "lr": 3e-05, "grad_norm": 0.318668653213803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132568963580, "tokens_per_second": 152217.31002958657, "data_wait_s": 1.2447999324649572e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905800, "loss": 0.5287038683891296, "lr": 3e-05, "grad_norm": 0.3419031326434391, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132583377866, "tokens_per_second": 150392.5037710462, "data_wait_s": 1.4029999874765053e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 905900, "loss": 0.54401695728302, "lr": 3e-05, "grad_norm": 0.3293203718565147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132597881696, "tokens_per_second": 151061.0973688057, "data_wait_s": 1.6254001820925623e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bdf17ca5bed4abf91e3f7b3ceb8677cf0f2145ef761e36a0c838bece80113146
-size 443841
+oid sha256:57e462d0025133afb6f15fec4602aa207dd386de536c1d92481b1f665d7d2219
+size 445959