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": 76732.33442309573,
-      "required_pages_per_second_upper_estimate": 0.0053085209389352375
+      "required_token_positions_per_second": 76838.65640177508,
+      "required_pages_per_second_upper_estimate": 0.005315876540121255
     },
     {
       "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": 9591.541802886966,
-      "required_pages_per_second_upper_estimate": 0.004015778258699719
+      "required_token_positions_per_second": 9604.832050221885,
+      "required_pages_per_second_upper_estimate": 0.004021342607726926
     },
     {
       "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": 19183.083605773933,
-      "required_pages_per_second_upper_estimate": 0.0039953304231482665
+      "required_token_positions_per_second": 19209.66410044377,
+      "required_pages_per_second_upper_estimate": 0.0040008664392132106
     },
     {
       "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": 28774.625408660897,
-      "required_pages_per_second_upper_estimate": 0.0017249948569402166
+      "required_token_positions_per_second": 28814.496150665655,
+      "required_pages_per_second_upper_estimate": 0.0017273850470443037
     },
     {
       "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": 28774.625408660897,
-      "required_pages_per_second_upper_estimate": 0.0022371005593221404
+      "required_token_positions_per_second": 28814.496150665655,
+      "required_pages_per_second_upper_estimate": 0.0022402003341401504
     },
     {
       "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": 28774.625408660897,
-      "required_pages_per_second_upper_estimate": 0.36908999895666933
+      "required_token_positions_per_second": 28814.496150665655,
+      "required_pages_per_second_upper_estimate": 0.36960141802523894
     }
   ],
   "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": 0.05748125912506907,
-    "producer_p95_seconds": 0.18829417199958698,
-    "gpu_update_seconds": 0.9608465655882424,
-    "producer_batches_per_second": 17.39697451345275,
-    "gpu_batches_per_second": 1.0407488935424225,
-    "producer_headroom": 16.71582321287726,
+    "producer_batch_seconds": 0.10905349553907229,
+    "producer_p95_seconds": 0.2027186380000785,
+    "gpu_update_seconds": 0.9595170380711756,
+    "producer_batches_per_second": 9.169811522838481,
+    "gpu_batches_per_second": 1.0421909776716456,
+    "producer_headroom": 8.798590392064916,
     "sustainable": true,
-    "buffer_seconds": 60.533333632059275,
+    "buffer_seconds": 60.44957339848406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295428.87405359914,
-    "upload_bytes_per_second": 5504903.274550938,
+    "generation_bytes_per_second": 319320.5037880451,
+    "upload_bytes_per_second": 5518920.010587513,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.392159139690214,
+    "estimated_full_shards_per_hour": 68.51874671202674,
     "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": 1.0483822090000103,
-      "bytes": 6048562,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1226819749999777,
       "bytes": 7652419,
@@ -562,6 +556,12 @@
       "bytes": 1047345,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0591198959991743,
+      "bytes": 7189815,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 615100, "loss": 0.5733166933059692, "lr": 0.0001, "grad_norm": 0.17269740992149213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90032860409, "tokens_per_second": 135606.47045930635, "data_wait_s": 1.533499926154036e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 615200, "loss": 0.5814242362976074, "lr": 0.0001, "grad_norm": 0.15711200904140998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90047330247, "tokens_per_second": 150724.3379113188, "data_wait_s": 6.053998731658794e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 615300, "loss": 0.54539555311203, "lr": 0.0001, "grad_norm": 0.16012991082265346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90061832092, "tokens_per_second": 151159.41338497744, "data_wait_s": 1.398799940943718e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 615400, "loss": 0.6175241470336914, "lr": 0.0001, "grad_norm": 0.179007262244534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90076518245, "tokens_per_second": 152883.73590053653, "data_wait_s": 1.0615000064717606e-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 sha256:9e151d201d87888bd62e5649602b3e58ef15c1150dd6f71abe65e4cfb05f2fcb
-size 233376
+oid sha256:e39d5de59e5ecfbf7d9e3b9650a5cb4b38a0fee5aa91ab9d7c742193acc82791
+size 235494