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": 76317.27781508541,
-      "required_pages_per_second_upper_estimate": 0.005279806359729069
+      "required_token_positions_per_second": 77234.30730638928,
+      "required_pages_per_second_upper_estimate": 0.005343248587739044
     },
     {
       "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": 9539.659726885677,
-      "required_pages_per_second_upper_estimate": 0.0039940563168989265
+      "required_token_positions_per_second": 9654.28841329866,
+      "required_pages_per_second_upper_estimate": 0.004042048954180871
     },
     {
       "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": 19079.319453771353,
-      "required_pages_per_second_upper_estimate": 0.003973719086730836
+      "required_token_positions_per_second": 19308.57682659732,
+      "required_pages_per_second_upper_estimate": 0.0040214673515669925
     },
     {
       "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": 28618.979180657032,
-      "required_pages_per_second_upper_estimate": 0.0017156641032294142
+      "required_token_positions_per_second": 28962.865239895982,
+      "required_pages_per_second_upper_estimate": 0.0017362795473971802
     },
     {
       "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": 28618.979180657032,
-      "required_pages_per_second_upper_estimate": 0.0022249997497102445
+      "required_token_positions_per_second": 28962.865239895982,
+      "required_pages_per_second_upper_estimate": 0.0022517353782211544
     },
     {
       "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": 28618.979180657032,
-      "required_pages_per_second_upper_estimate": 0.36709353626373487
+      "required_token_positions_per_second": 28962.865239895982,
+      "required_pages_per_second_upper_estimate": 0.3715045373955693
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.31017560567185853,
-    "producer_p95_seconds": 0.1751804280002034,
-    "gpu_update_seconds": 0.9660721937519946,
-    "producer_batches_per_second": 3.2239801638621497,
-    "gpu_batches_per_second": 1.0351193280040882,
-    "producer_headroom": 3.114597589515222,
+    "producer_batch_seconds": 0.0552023849297143,
+    "producer_p95_seconds": 0.16888296599972819,
+    "gpu_update_seconds": 0.9546016863661413,
+    "producer_batches_per_second": 18.115159359024737,
+    "gpu_batches_per_second": 1.0475573365124415,
+    "producer_headroom": 17.2927616729164,
     "sustainable": true,
-    "buffer_seconds": 60.862548206375656,
+    "buffer_seconds": 60.1399062410669,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411540.51925968047,
-    "upload_bytes_per_second": 5008877.156159869,
+    "generation_bytes_per_second": 412221.22224988433,
+    "upload_bytes_per_second": 5011068.203523412,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.307015260151,
+    "estimated_full_shards_per_hour": 88.45307827589414,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 13455700,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.9797629530003178,
+      "bytes": 4961107,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 765600, "loss": 0.5014462471008301, "lr": 3e-05, "grad_norm": 0.23670771429161894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112061341022, "tokens_per_second": 153356.97703893794, "data_wait_s": 1.1054000424337573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765700, "loss": 0.5179704427719116, "lr": 3e-05, "grad_norm": 0.2353772944463407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112076022277, "tokens_per_second": 152006.66279545918, "data_wait_s": 1.2077000064891763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765800, "loss": 0.5298739671707153, "lr": 3e-05, "grad_norm": 0.24303016357071958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112090777686, "tokens_per_second": 153162.50336761388, "data_wait_s": 1.949299985426478e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 765900, "loss": 0.5385651588439941, "lr": 3e-05, "grad_norm": 0.24259133436256383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112105413335, "tokens_per_second": 151001.37321612364, "data_wait_s": 1.3341999874683097e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:677f752995e0581907778a6f33d10b23d5ae5cecdeab0c8fc473351138842831
-size 123792
+oid sha256:f1a6878d044b62c2f3b263d4a72f1acd88460913a9f626ab4af02a9a4ef0b6d4
+size 125910