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": 77124.95304100928,
-      "required_pages_per_second_upper_estimate": 0.005335683205923202
+      "required_token_positions_per_second": 77250.65320733543,
+      "required_pages_per_second_upper_estimate": 0.005344379435094215
     },
     {
       "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": 9640.61913012616,
-      "required_pages_per_second_upper_estimate": 0.004036325910763638
+      "required_token_positions_per_second": 9656.331650916929,
+      "required_pages_per_second_upper_estimate": 0.004042904415103987
     },
     {
       "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": 19281.23826025232,
-      "required_pages_per_second_upper_estimate": 0.004015773449163806
+      "required_token_positions_per_second": 19312.663301833858,
+      "required_pages_per_second_upper_estimate": 0.004022318456591165
     },
     {
       "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": 28921.857390378485,
-      "required_pages_per_second_upper_estimate": 0.0017338211894339668
+      "required_token_positions_per_second": 28968.99495275079,
+      "required_pages_per_second_upper_estimate": 0.0017366470143232963
     },
     {
       "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": 28921.857390378485,
-      "required_pages_per_second_upper_estimate": 0.0022485471982956365
+      "required_token_positions_per_second": 28968.99495275079,
+      "required_pages_per_second_upper_estimate": 0.002252211936433858
     },
     {
       "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": 28921.857390378485,
-      "required_pages_per_second_upper_estimate": 0.3709785327327572
+      "required_token_positions_per_second": 28968.99495275079,
+      "required_pages_per_second_upper_estimate": 0.3715831627704979
     }
   ],
   "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": 0.26301998225784473,
-    "producer_p95_seconds": 0.2641290920000756,
-    "gpu_update_seconds": 0.9559552011759016,
-    "producer_batches_per_second": 3.801992500401267,
-    "gpu_batches_per_second": 1.0460741243626477,
-    "producer_headroom": 3.6345345055903624,
+    "producer_batch_seconds": 0.22385546586718874,
+    "producer_p95_seconds": 0.2586002800017013,
+    "gpu_update_seconds": 0.9543996968170498,
+    "producer_batches_per_second": 4.467168117276574,
+    "gpu_batches_per_second": 1.0477790419831738,
+    "producer_headroom": 4.263463896759554,
     "sustainable": true,
-    "buffer_seconds": 60.2251776740818,
+    "buffer_seconds": 60.12718089947414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 536067.075408576,
-    "upload_bytes_per_second": 4932858.767119787,
+    "generation_bytes_per_second": 513341.94122891984,
+    "upload_bytes_per_second": 4886559.329300388,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.02751537983858,
+    "estimated_full_shards_per_hour": 110.15123059893318,
     "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": 1.363758608000353,
-      "bytes": 11974676,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 1.1600040840003203,
       "bytes": 6119662,
@@ -562,6 +556,12 @@
       "bytes": 11400278,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.142467518999183,
+      "bytes": 6818785,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -223,3 +223,4 @@
 {"step": 907300, "loss": 0.5100760459899902, "lr": 3e-05, "grad_norm": 0.3225271213909381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132802535898, "tokens_per_second": 152268.30793594522, "data_wait_s": 1.2900996807729825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907400, "loss": 0.509017288684845, "lr": 3e-05, "grad_norm": 0.3516123061616303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132817226912, "tokens_per_second": 152125.18411334202, "data_wait_s": 1.1537002137629315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907500, "loss": 0.5206209421157837, "lr": 3e-05, "grad_norm": 0.32198816843496814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132831872870, "tokens_per_second": 152057.82358131764, "data_wait_s": 1.4343000657390803e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 907600, "loss": 0.5512479543685913, "lr": 3e-05, "grad_norm": 0.3486921869976473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132846430728, "tokens_per_second": 152014.17575299123, "data_wait_s": 1.1598996934480965e-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 sha256:60b16af38ffcc9149f31f9adc4dba586088d79423922ae4cd941eb1c253c1dc6
-size 479847
+oid sha256:24dbf30548aab60116ad37eb5892a0f44d7c0ff08b0b0ec09a08d034dc39b0b5
+size 481965