Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74659.88998972389,
-      "required_pages_per_second_upper_estimate": 0.006471330040548055
+      "required_token_positions_per_second": 74569.75015202582,
+      "required_pages_per_second_upper_estimate": 0.006463516947873717
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9332.486248715486,
-      "required_pages_per_second_upper_estimate": 0.0037228682977164056
+      "required_token_positions_per_second": 9321.218769003228,
+      "required_pages_per_second_upper_estimate": 0.0037183735315953516
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18664.972497430972,
-      "required_pages_per_second_upper_estimate": 0.004317812894984106
+      "required_token_positions_per_second": 18642.437538006456,
+      "required_pages_per_second_upper_estimate": 0.004312599828722984
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27997.458746146458,
-      "required_pages_per_second_upper_estimate": 0.0017048983463266371
+      "required_token_positions_per_second": 27963.65630700968,
+      "required_pages_per_second_upper_estimate": 0.0017028399551308995
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27997.458746146458,
-      "required_pages_per_second_upper_estimate": 0.002155321876204508
+      "required_token_positions_per_second": 27963.65630700968,
+      "required_pages_per_second_upper_estimate": 0.0021527196708686185
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27997.458746146458,
-      "required_pages_per_second_upper_estimate": 0.34763473616035434
+      "required_token_positions_per_second": 27963.65630700968,
+      "required_pages_per_second_upper_estimate": 0.34721502299576196
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10644786699998576,
-    "producer_p95_seconds": 0.22018297300019185,
-    "gpu_update_seconds": 0.9875181976580443,
-    "producer_batches_per_second": 9.394269966913793,
-    "gpu_batches_per_second": 1.0126395669179131,
-    "producer_headroom": 9.277012546039803,
+    "producer_batch_seconds": 0.074832562304735,
+    "producer_p95_seconds": 0.19214319799903024,
+    "gpu_update_seconds": 0.988711908645131,
+    "producer_batches_per_second": 13.363166637643321,
+    "gpu_batches_per_second": 1.011416967122746,
+    "producer_headroom": 13.212321991847265,
     "sustainable": true,
-    "buffer_seconds": 60.238610057140704,
+    "buffer_seconds": 60.31142642735299,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313868.1640997037,
-    "upload_bytes_per_second": 4006655.738620722,
+    "generation_bytes_per_second": 299588.0660923766,
+    "upload_bytes_per_second": 4058653.93467418,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.34880153888066,
+    "estimated_full_shards_per_hour": 64.28462493017648,
     "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": 0.911286476999976,
-      "bytes": 856023,
-      "pages": 24,
-      "shards": 1
-    },
     {
       "seconds": 1.8453115359998264,
       "bytes": 9879354,
@@ -562,6 +556,12 @@
       "bytes": 1115458,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1859958829991228,
+      "bytes": 7375397,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -97,3 +97,4 @@
 {"step": 324700, "loss": 0.5599087551236153, "lr": 0.0003, "grad_norm": 0.1449137258835772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47526927889, "tokens_per_second": 148547.9794566315, "data_wait_s": 1.4236999049899168e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324800, "loss": 0.5679438281804323, "lr": 0.0003, "grad_norm": 0.1692994539482135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47541583873, "tokens_per_second": 149203.58725814408, "data_wait_s": 1.4347999240271747e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 324900, "loss": 0.5770327933132648, "lr": 0.0003, "grad_norm": 0.15126750884586188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47556306635, "tokens_per_second": 150053.71359794465, "data_wait_s": 1.119599983212538e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325000, "loss": 0.6160995494574308, "lr": 0.0003, "grad_norm": 0.14224377468636584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47571096641, "tokens_per_second": 151367.18161095746, "data_wait_s": 1.3118000424583443e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13367d03513b1ad74cebd0ffd97e3ed345a1e771636abb4d49bb10d09eba2a4f
-size 210630
+oid sha256:75d8ba9cea74193dff7bb7e5c51a23369b9596aad91572b432a8b1d586e05f9d
+size 212748