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": 74525.46017424166,
-      "required_pages_per_second_upper_estimate": 0.006459678004851308
+      "required_token_positions_per_second": 75259.72559165455,
+      "required_pages_per_second_upper_estimate": 0.006523322270253972
     },
     {
       "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": 9315.682521780207,
-      "required_pages_per_second_upper_estimate": 0.0037161650398038165
+      "required_token_positions_per_second": 9407.46569895682,
+      "required_pages_per_second_upper_estimate": 0.0037527787214603554
     },
     {
       "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": 18631.365043560414,
-      "required_pages_per_second_upper_estimate": 0.004310038402002139
+      "required_token_positions_per_second": 18814.93139791364,
+      "required_pages_per_second_upper_estimate": 0.004352503247424261
     },
     {
       "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": 27947.047565340617,
-      "required_pages_per_second_upper_estimate": 0.0017018285699025866
+      "required_token_positions_per_second": 28222.39709687046,
+      "required_pages_per_second_upper_estimate": 0.0017185959117254103
     },
     {
       "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": 27947.047565340617,
-      "required_pages_per_second_upper_estimate": 0.0021514410839589943
+      "required_token_positions_per_second": 28222.39709687046,
+      "required_pages_per_second_upper_estimate": 0.0021726382531124487
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27947.047565340617,
-      "required_pages_per_second_upper_estimate": 0.3470087980101148
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28222.39709687046,
+      "required_pages_per_second_upper_estimate": 0.3504277176561141
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13161937139071256,
-    "producer_p95_seconds": 0.3195129529995029,
-    "gpu_update_seconds": 0.9892994934566365,
-    "producer_batches_per_second": 7.597665825583504,
-    "gpu_batches_per_second": 1.0108162458528869,
-    "producer_headroom": 7.516366952702558,
+    "producer_batch_seconds": 0.2116494245310605,
+    "producer_p95_seconds": 0.3471123869967414,
+    "gpu_update_seconds": 0.9796474730725777,
+    "producer_batches_per_second": 4.724794325406944,
+    "gpu_batches_per_second": 1.0207753579597245,
+    "producer_headroom": 4.628632821672568,
     "sustainable": true,
-    "buffer_seconds": 60.347269100854824,
+    "buffer_seconds": 60.73814333049982,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535948.6288296074,
-    "upload_bytes_per_second": 3543240.729573944,
+    "generation_bytes_per_second": 513340.5743547225,
+    "upload_bytes_per_second": 3617120.5429235776,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.00209950128713,
+    "estimated_full_shards_per_hour": 110.15093729954964,
     "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": 0.9514959919997636,
-      "bytes": 336067,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 1.108248578999337,
       "bytes": 861803,
@@ -562,6 +556,12 @@
       "bytes": 8475194,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.4765983690012945,
+      "bytes": 10921563,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 217700, "loss": 0.5840618945658207, "lr": 0.0003, "grad_norm": 0.14450991405010163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31864315380, "tokens_per_second": 148098.04722320035, "data_wait_s": 1.3136999768903479e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217800, "loss": 0.5621696505695581, "lr": 0.0003, "grad_norm": 0.1469949327678859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31878829543, "tokens_per_second": 147603.85871186946, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217900, "loss": 0.6120729297399521, "lr": 0.0003, "grad_norm": 0.1569615759045584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31893514828, "tokens_per_second": 149033.36150266792, "data_wait_s": 1.341800088994205e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 218000, "loss": 0.606307540088892, "lr": 0.0003, "grad_norm": 0.14625266571941004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31908214211, "tokens_per_second": 148565.29103573423, "data_wait_s": 1.2604999938048422e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1929dc9cacd8c4d19c4a1c5accf4b3d3fbbfb001eb132928cb1b214e5c3ea8e
-size 488319
+oid sha256:7059915a143ed41b9e68ff51440bc0540ee1c2d51c054288d999adeda5211451
+size 490437