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": 75637.67374212216,
-      "required_pages_per_second_upper_estimate": 0.006556081858035674
+      "required_token_positions_per_second": 75665.89182128214,
+      "required_pages_per_second_upper_estimate": 0.006558527729619193
     },
     {
       "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": 9454.70921776527,
-      "required_pages_per_second_upper_estimate": 0.003771624867466599
+      "required_token_positions_per_second": 9458.236477660268,
+      "required_pages_per_second_upper_estimate": 0.003773031944175949
     },
     {
       "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": 18909.41843553054,
-      "required_pages_per_second_upper_estimate": 0.004374361160661865
+      "required_token_positions_per_second": 18916.472955320536,
+      "required_pages_per_second_upper_estimate": 0.004375993099659973
     },
     {
       "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": 28364.127653295807,
-      "required_pages_per_second_upper_estimate": 0.001727226559008953
+      "required_token_positions_per_second": 28374.709432980802,
+      "required_pages_per_second_upper_estimate": 0.001727870933873462
     },
     {
       "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": 28364.127653295807,
-      "required_pages_per_second_upper_estimate": 0.0021835490636813626
+      "required_token_positions_per_second": 28374.709432980802,
+      "required_pages_per_second_upper_estimate": 0.0021843636783737547
     },
     {
       "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": 28364.127653295807,
-      "required_pages_per_second_upper_estimate": 0.35218753682525805
+      "required_token_positions_per_second": 28374.709432980802,
+      "required_pages_per_second_upper_estimate": 0.35231892711400725
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10777247354698716,
-    "producer_p95_seconds": 0.23451162899982592,
-    "gpu_update_seconds": 0.9747523469768126,
-    "producer_batches_per_second": 9.278807167435154,
-    "gpu_batches_per_second": 1.0259016078304328,
-    "producer_headroom": 9.044539063602686,
+    "producer_batch_seconds": 0.5762099086640688,
+    "producer_p95_seconds": 0.2648093690004316,
+    "gpu_update_seconds": 0.9743888326082336,
+    "producer_batches_per_second": 1.7354786597101046,
+    "gpu_batches_per_second": 1.0262843400239006,
+    "producer_headroom": 1.6910310252514307,
     "sustainable": true,
-    "buffer_seconds": 60.434645512562376,
+    "buffer_seconds": 60.41210762171048,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359453.47249555204,
-    "upload_bytes_per_second": 5509164.752400412,
+    "generation_bytes_per_second": 316487.0910617126,
+    "upload_bytes_per_second": 5606617.833600164,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.13034754896088,
+    "estimated_full_shards_per_hour": 67.9107622994283,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 927364,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.263599650001197,
+      "bytes": 13782793,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 287300, "loss": 0.6012470815330744, "lr": 0.0003, "grad_norm": 0.13943662973821655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42052528223, "tokens_per_second": 151016.36593036633, "data_wait_s": 1.1636999261099845e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287400, "loss": 0.6403681021183729, "lr": 0.0003, "grad_norm": 0.15959914732462238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42067193688, "tokens_per_second": 149597.78222870533, "data_wait_s": 1.1025000276276842e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287500, "loss": 0.6373869748786092, "lr": 0.0003, "grad_norm": 0.17657408476356823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42081977746, "tokens_per_second": 151614.5092031017, "data_wait_s": 1.4546001693815924e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 287600, "loss": 0.5928755290806293, "lr": 0.0003, "grad_norm": 0.15635010854087456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42096763738, "tokens_per_second": 125391.40559606963, "data_wait_s": 8.151000656653196e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:169c47876600a9f913c980eeb85b23fde4e7a81cdefc1df84a622047019abdad
-size 155562
+oid sha256:5c2552bad5f4275a3386df000873ef6e1c507c72c8158de65b99d1f179e1be45
+size 157680