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": 74362.84339863072,
-      "required_pages_per_second_upper_estimate": 0.006445582794889802
+      "required_token_positions_per_second": 74707.1110467094,
+      "required_pages_per_second_upper_estimate": 0.0064754230420868844
     },
     {
       "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": 9295.35542482884,
-      "required_pages_per_second_upper_estimate": 0.003708056256912733
+      "required_token_positions_per_second": 9338.388880838675,
+      "required_pages_per_second_upper_estimate": 0.0037252229459225606
     },
     {
       "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": 18590.71084965768,
-      "required_pages_per_second_upper_estimate": 0.004300633769732117
+      "required_token_positions_per_second": 18676.77776167735,
+      "required_pages_per_second_upper_estimate": 0.004320543835101951
     },
     {
       "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": 27886.066274486522,
-      "required_pages_per_second_upper_estimate": 0.0016981151292336787
+      "required_token_positions_per_second": 28015.166642516022,
+      "required_pages_per_second_upper_estimate": 0.0017059766643093915
     },
     {
       "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": 27886.066274486522,
-      "required_pages_per_second_upper_estimate": 0.0021467465753820286
+      "required_token_positions_per_second": 28015.166642516022,
+      "required_pages_per_second_upper_estimate": 0.002156685079085631
     },
     {
       "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": 27886.066274486522,
-      "required_pages_per_second_upper_estimate": 0.34625161446895864
+      "required_token_positions_per_second": 28015.166642516022,
+      "required_pages_per_second_upper_estimate": 0.34785460896874754
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09248980873439905,
-    "producer_p95_seconds": 0.30514711100113345,
-    "gpu_update_seconds": 0.9914628950478995,
-    "producer_batches_per_second": 10.812002032263663,
-    "gpu_batches_per_second": 1.0086106146732683,
-    "producer_headroom": 10.719698836171904,
+    "producer_batch_seconds": 0.2887545934531772,
+    "producer_p95_seconds": 0.2855890040009399,
+    "gpu_update_seconds": 0.9868940046939143,
+    "producer_batches_per_second": 3.463148371221164,
+    "gpu_batches_per_second": 1.0132800434937799,
+    "producer_headroom": 3.417760364923661,
     "sustainable": true,
-    "buffer_seconds": 60.47923659792187,
+    "buffer_seconds": 60.200534286328775,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 583952.0878516773,
+    "generation_bytes_per_second": 540630.5717055602,
     "upload_bytes_per_second": 3580847.63563265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.65126217204447,
+    "estimated_full_shards_per_hour": 116.00673545241456,
     "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": [

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

@@ -225,3 +225,4 @@
 {"step": 217500, "loss": 0.6024927664548159, "lr": 0.0003, "grad_norm": 0.158279336983577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31835045908, "tokens_per_second": 148549.87263134716, "data_wait_s": 1.2456999684218317e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217600, "loss": 0.6607991196215153, "lr": 0.0003, "grad_norm": 0.18329649626859376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31849671581, "tokens_per_second": 148777.65195918386, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:36619bf57ee1e1567da417fc5b9af4dbb52858310e633e53563b71c3827c7082
-size 484083
+oid sha256:5069d7068b8048c4e1689824cff7e1b81a5b672947aac10b9fc2db4c0495df82
+size 486201