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": 75551.06748499216,
-      "required_pages_per_second_upper_estimate": 0.006548575047169197
+      "required_token_positions_per_second": 75201.92783982909,
+      "required_pages_per_second_upper_estimate": 0.006518312507612803
     },
     {
       "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": 9443.88343562402,
-      "required_pages_per_second_upper_estimate": 0.003767306301110587
+      "required_token_positions_per_second": 9400.240979978636,
+      "required_pages_per_second_upper_estimate": 0.0037498966730407837
     },
     {
       "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": 18887.76687124804,
-      "required_pages_per_second_upper_estimate": 0.004369352452319625
+      "required_token_positions_per_second": 18800.481959957273,
+      "required_pages_per_second_upper_estimate": 0.004349160624254479
     },
     {
       "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": 28331.650306872063,
-      "required_pages_per_second_upper_estimate": 0.0017252488590072147
+      "required_token_positions_per_second": 28200.72293993591,
+      "required_pages_per_second_upper_estimate": 0.001717276069283609
     },
     {
       "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": 28331.650306872063,
-      "required_pages_per_second_upper_estimate": 0.002181048867650614
+      "required_token_positions_per_second": 28200.72293993591,
+      "required_pages_per_second_upper_estimate": 0.0021709697165137208
     },
     {
       "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": 28331.650306872063,
-      "required_pages_per_second_upper_estimate": 0.3517842768773615
+      "required_token_positions_per_second": 28200.72293993591,
+      "required_pages_per_second_upper_estimate": 0.35015859716572395
     }
   ],
   "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": 0.0969070824689311,
-    "producer_p95_seconds": 0.30215932999999495,
-    "gpu_update_seconds": 0.975869732279371,
-    "producer_batches_per_second": 10.319163207917287,
-    "gpu_batches_per_second": 1.0247269352890649,
-    "producer_headroom": 10.070159037057378,
+    "producer_batch_seconds": 0.7014030111173497,
+    "producer_p95_seconds": 0.31772967700089794,
+    "gpu_update_seconds": 0.9804003982056367,
+    "producer_batches_per_second": 1.4257138679900718,
+    "gpu_batches_per_second": 1.019991425778932,
+    "producer_headroom": 1.3977704439047651,
     "sustainable": true,
-    "buffer_seconds": 60.503923401321,
+    "buffer_seconds": 60.78482468874948,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363164.37556073116,
-    "upload_bytes_per_second": 6019438.499153285,
+    "generation_bytes_per_second": 311911.1720628726,
+    "upload_bytes_per_second": 5907427.702888135,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.92662096134616,
+    "estimated_full_shards_per_hour": 66.92887660421977,
     "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": 1.2326910120027605,
-      "bytes": 11762073,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 0.9280554619981558,
       "bytes": 1182136,
@@ -562,6 +556,12 @@
       "bytes": 6252756,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9329704880001373,
+      "bytes": 2137944,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -161,3 +161,4 @@
 {"step": 101100, "loss": 0.6251995395869017, "lr": 0.0003, "grad_norm": 0.13753801862563045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14794354040, "tokens_per_second": 149332.41022912136, "data_wait_s": 8.861999958753586e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 101200, "loss": 0.6119210012257099, "lr": 0.0003, "grad_norm": 0.12931871012234633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14809187848, "tokens_per_second": 151153.94490864305, "data_wait_s": 1.4155997632769868e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101300, "loss": 0.5754358042031527, "lr": 0.0003, "grad_norm": 0.17327013163768856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14823943277, "tokens_per_second": 148975.148350734, "data_wait_s": 2.0988998585380614e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 101400, "loss": 0.597405731678009, "lr": 0.0003, "grad_norm": 0.16215826759214721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14838427351, "tokens_per_second": 147424.06715320845, "data_wait_s": 1.7465998098487034e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2abcdf6830fc379792af5daec282da2d84d1fa8397f2cef5852cce8ae8427432
-size 347748
+oid sha256:5732a5bdf65f791e4b74ab7b4b427035ee5dcf90136716abe2a74558c2355e0b
+size 349866