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": 76261.59023609251,
-      "required_pages_per_second_upper_estimate": 0.0052759537638013545
+      "required_token_positions_per_second": 76109.20646889936,
+      "required_pages_per_second_upper_estimate": 0.005265411501207878
     },
     {
       "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": 9532.698779511564,
-      "required_pages_per_second_upper_estimate": 0.003991141913594496
+      "required_token_positions_per_second": 9513.65080861242,
+      "required_pages_per_second_upper_estimate": 0.003983166925945888
     },
     {
       "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": 19065.397559023128,
-      "required_pages_per_second_upper_estimate": 0.003970819523199835
+      "required_token_positions_per_second": 19027.30161722484,
+      "required_pages_per_second_upper_estimate": 0.003962885143180796
     },
     {
       "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": 28598.09633853469,
-      "required_pages_per_second_upper_estimate": 0.0017144122087304336
+      "required_token_positions_per_second": 28540.95242583726,
+      "required_pages_per_second_upper_estimate": 0.001710986518417925
     },
     {
       "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": 28598.09633853469,
-      "required_pages_per_second_upper_estimate": 0.0022233762005891477
+      "required_token_positions_per_second": 28540.95242583726,
+      "required_pages_per_second_upper_estimate": 0.002218933512726431
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28598.09633853469,
-      "required_pages_per_second_upper_estimate": 0.3668256735872384
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28540.95242583726,
+      "required_pages_per_second_upper_estimate": 0.3660926928315088
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.24594400135941896,
-    "producer_p95_seconds": 0.17745527500028402,
-    "gpu_update_seconds": 0.9667776369696861,
-    "producer_batches_per_second": 4.06596621374235,
-    "gpu_batches_per_second": 1.0343640168740846,
-    "producer_headroom": 3.93088520812041,
+    "producer_batch_seconds": 0.05334846459371789,
+    "producer_p95_seconds": 0.17214922000039223,
+    "gpu_update_seconds": 0.9687132926570139,
+    "producer_batches_per_second": 18.744681925068114,
+    "gpu_batches_per_second": 1.0322971797539517,
+    "producer_headroom": 18.158222547441145,
     "sustainable": true,
-    "buffer_seconds": 60.90699112909022,
+    "buffer_seconds": 60.06022414473486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 530631.7793589398,
-    "upload_bytes_per_second": 5355182.496344951,
+    "generation_bytes_per_second": 529997.9433788116,
+    "upload_bytes_per_second": 5399154.946430673,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.86122737480304,
+    "estimated_full_shards_per_hour": 113.72522092841398,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 5407000,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0307093159999567,
+      "bytes": 7655434,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -47,3 +47,4 @@
 {"step": 609100, "loss": 0.5480009913444519, "lr": 0.0001, "grad_norm": 0.13478104810810157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89154716094, "tokens_per_second": 153392.15708586172, "data_wait_s": 1.6027000128815416e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609200, "loss": 0.5573991537094116, "lr": 0.0001, "grad_norm": 0.16132565392282786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89169170728, "tokens_per_second": 150233.63629403894, "data_wait_s": 2.0789999325643294e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609300, "loss": 0.674561083316803, "lr": 0.0001, "grad_norm": 0.1675273860365766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89183689167, "tokens_per_second": 149751.65645260023, "data_wait_s": 1.0541999472479802e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 609400, "loss": 0.527541995048523, "lr": 0.0001, "grad_norm": 0.15548710990944486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89198141513, "tokens_per_second": 149871.48394250727, "data_wait_s": 1.3796000530419406e-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 sha256:1087e7f3526cebc660c645e57d500fe3ff80a9af0ae698daa7e8c147243ee048
-size 104730
+oid sha256:60594d8ab70197ee6c146c6d9a9243b30167e04229ad7cba7b44904c3afe3b5a
+size 106848