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": 75259.72559165455,
-      "required_pages_per_second_upper_estimate": 0.006523322270253972
+      "required_token_positions_per_second": 75459.5416336954,
+      "required_pages_per_second_upper_estimate": 0.006540641818348943
     },
     {
       "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": 9407.46569895682,
-      "required_pages_per_second_upper_estimate": 0.0037527787214603554
+      "required_token_positions_per_second": 9432.442704211924,
+      "required_pages_per_second_upper_estimate": 0.003762742422296124
     },
     {
       "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": 18814.93139791364,
-      "required_pages_per_second_upper_estimate": 0.004352503247424261
+      "required_token_positions_per_second": 18864.88540842385,
+      "required_pages_per_second_upper_estimate": 0.004364059228595241
     },
     {
       "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": 28222.39709687046,
-      "required_pages_per_second_upper_estimate": 0.0017185959117254103
+      "required_token_positions_per_second": 28297.32811263577,
+      "required_pages_per_second_upper_estimate": 0.0017231588174528606
     },
     {
       "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": 28222.39709687046,
-      "required_pages_per_second_upper_estimate": 0.0021726382531124487
+      "required_token_positions_per_second": 28297.32811263577,
+      "required_pages_per_second_upper_estimate": 0.0021784066501283894
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28222.39709687046,
-      "required_pages_per_second_upper_estimate": 0.3504277176561141
+      "required_token_positions_per_second": 28297.32811263577,
+      "required_pages_per_second_upper_estimate": 0.3513581100939415
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.09865101757813477,
+    "producer_p95_seconds": 0.3212924279978324,
+    "gpu_update_seconds": 0.9770533772640597,
+    "producer_batches_per_second": 10.136742879595417,
+    "gpu_batches_per_second": 1.0234855364813285,
+    "producer_headroom": 9.904138864966113,
     "sustainable": true,
-    "buffer_seconds": 60.73814333049982,
+    "buffer_seconds": 60.5773093903717,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 513340.5743547225,
-    "upload_bytes_per_second": 3617120.5429235776,
+    "generation_bytes_per_second": 526154.4247739561,
+    "upload_bytes_per_second": 3627485.5486378567,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.15093729954964,
+    "estimated_full_shards_per_hour": 112.9004913083459,
     "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": 1.108248578999337,
-      "bytes": 861803,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.207310141000562,
       "bytes": 13710377,
@@ -562,6 +556,12 @@
       "bytes": 10921563,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.8757540010010416,
+      "bytes": 1242501,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -228,3 +228,4 @@
 {"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}
+{"step": 218100, "loss": 0.6077547408640385, "lr": 0.0003, "grad_norm": 0.1613877905739051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31922814983, "tokens_per_second": 147908.95578827965, "data_wait_s": 1.1562999134184793e-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 sha256:7059915a143ed41b9e68ff51440bc0540ee1c2d51c054288d999adeda5211451
-size 490437
+oid sha256:a5aaa71072fcb24b20c56b365ef8c37d01acc073971f78594ccf16fad9dd914f
+size 492555