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": 74600.49634397197,
-      "required_pages_per_second_upper_estimate": 0.006466181949866099
+      "required_token_positions_per_second": 76228.63617593308,
+      "required_pages_per_second_upper_estimate": 0.006607304984017809
     },
     {
       "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": 9325.062042996497,
-      "required_pages_per_second_upper_estimate": 0.0037199066710533334
+      "required_token_positions_per_second": 9528.579521991634,
+      "required_pages_per_second_upper_estimate": 0.0038010928362819667
     },
     {
       "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": 18650.124085992993,
-      "required_pages_per_second_upper_estimate": 0.004314377976321045
+      "required_token_positions_per_second": 19057.15904398327,
+      "required_pages_per_second_upper_estimate": 0.004408538350281501
     },
     {
       "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": 27975.18612898949,
-      "required_pages_per_second_upper_estimate": 0.0017035420607971691
+      "required_token_positions_per_second": 28585.7385659749,
+      "required_pages_per_second_upper_estimate": 0.0017407215008885093
     },
     {
       "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": 27975.18612898949,
-      "required_pages_per_second_upper_estimate": 0.002153607268481214
+      "required_token_positions_per_second": 28585.7385659749,
+      "required_pages_per_second_upper_estimate": 0.002200609285197669
     },
     {
       "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": 27975.18612898949,
-      "required_pages_per_second_upper_estimate": 0.3473581847969193
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28585.7385659749,
+      "required_pages_per_second_upper_estimate": 0.3549392026767188
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15212802719537422,
-    "producer_p95_seconds": 0.39954024799953913,
-    "gpu_update_seconds": 0.9883044163681026,
-    "producer_batches_per_second": 6.573410688588797,
-    "gpu_batches_per_second": 1.0118339890404187,
-    "producer_headroom": 6.496530814133598,
+    "producer_batch_seconds": 0.42674889275770056,
+    "producer_p95_seconds": 0.40972126799897524,
+    "gpu_update_seconds": 0.9671955802782344,
+    "producer_batches_per_second": 2.3432984056218276,
+    "gpu_batches_per_second": 1.0339170488272174,
+    "producer_headroom": 2.266427861190465,
     "sustainable": true,
-    "buffer_seconds": 60.28656939845426,
+    "buffer_seconds": 60.933321557528764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435050.01589050767,
-    "upload_bytes_per_second": 5762266.3072296465,
+    "generation_bytes_per_second": 383030.5496539843,
+    "upload_bytes_per_second": 5746051.909421746,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.35160596405433,
+    "estimated_full_shards_per_hour": 82.18943946089408,
     "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": 0.946805650004535,
-      "bytes": 8294572,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.4035626420009066,
       "bytes": 10955791,
@@ -562,6 +556,12 @@
       "bytes": 1047459,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.009063662000699,
+      "bytes": 7464078,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -337,3 +337,4 @@
 {"step": 313900, "loss": 0.6181909441947937, "lr": 0.0003, "grad_norm": 0.15279418255568192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45946273386, "tokens_per_second": 150840.54126490458, "data_wait_s": 9.335002687294036e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 314000, "loss": 0.5206610346212983, "lr": 0.0003, "grad_norm": 0.15408984745557255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45961037178, "tokens_per_second": 151764.0794378983, "data_wait_s": 1.0324001777917147e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314100, "loss": 0.5576486960053444, "lr": 0.0003, "grad_norm": 0.15534584045263586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45975665453, "tokens_per_second": 149673.9852173171, "data_wait_s": 1.1534997611306608e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 314200, "loss": 0.5247049517929554, "lr": 0.0003, "grad_norm": 0.16029382198406003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45990174774, "tokens_per_second": 148771.5261618356, "data_wait_s": 1.0413998097646981e-05, "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 sha256:43ec1b0a2e3408e502c0e35a5e3237c5a11e30c980fc2eed15f1ec273481fa92
-size 722865
+oid sha256:5d67c4cb0c32522b7c9c60904bf0b585f8b01a2984ae545463d57520c7083b57
+size 724983