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": 74478.51791213546,
-      "required_pages_per_second_upper_estimate": 0.0064556091685460185
+      "required_token_positions_per_second": 75475.41268904982,
+      "required_pages_per_second_upper_estimate": 0.006542017481202242
     },
     {
       "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": 9309.814739016932,
-      "required_pages_per_second_upper_estimate": 0.0037138242935283758
+      "required_token_positions_per_second": 9434.426586131227,
+      "required_pages_per_second_upper_estimate": 0.0037635338224554123
     },
     {
       "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": 18619.629478033865,
-      "required_pages_per_second_upper_estimate": 0.004307323585456471
+      "required_token_positions_per_second": 18868.853172262454,
+      "required_pages_per_second_upper_estimate": 0.004364977100929044
     },
     {
       "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": 27929.444217050794,
-      "required_pages_per_second_upper_estimate": 0.001700756618349367
+      "required_token_positions_per_second": 28303.279758393677,
+      "required_pages_per_second_upper_estimate": 0.0017235212414536453
     },
     {
       "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": 27929.444217050794,
-      "required_pages_per_second_upper_estimate": 0.002150085929478457
+      "required_token_positions_per_second": 28303.279758393677,
+      "required_pages_per_second_upper_estimate": 0.002178864824294043
     },
     {
       "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": 27929.444217050794,
-      "required_pages_per_second_upper_estimate": 0.3467902233389721
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28303.279758393677,
+      "required_pages_per_second_upper_estimate": 0.35143200961537774
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10951564949216674,
-    "producer_p95_seconds": 0.2130781380001281,
-    "gpu_update_seconds": 0.989923028368786,
-    "producer_batches_per_second": 9.131115093021718,
-    "gpu_batches_per_second": 1.0101795506745803,
-    "producer_headroom": 9.039101105267989,
+    "producer_batch_seconds": 0.4853144441171935,
+    "producer_p95_seconds": 0.21884791900083655,
+    "gpu_update_seconds": 0.9768479213721036,
+    "producer_batches_per_second": 2.0605197560502044,
+    "gpu_batches_per_second": 1.0237008014465305,
+    "producer_headroom": 2.012814440643796,
     "sustainable": true,
-    "buffer_seconds": 60.385304730495946,
+    "buffer_seconds": 60.56457112507042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276429.9570833815,
-    "upload_bytes_per_second": 5859443.313682294,
+    "generation_bytes_per_second": 250000.7059543864,
+    "upload_bytes_per_second": 5805216.83603484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.31543383003314,
+    "estimated_full_shards_per_hour": 53.644331779229105,
     "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": 1.2221474409998336,
-      "bytes": 10132121,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.045252316999722,
       "bytes": 6879925,
@@ -562,6 +556,12 @@
       "bytes": 7722133,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0303949059998558,
+      "bytes": 5043636,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -136,3 +136,4 @@
 {"step": 368600, "loss": 0.5898146387189627, "lr": 0.0003, "grad_norm": 0.1585002514120512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53952706554, "tokens_per_second": 148790.51250604712, "data_wait_s": 1.0783000107039697e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368700, "loss": 0.562000397592783, "lr": 0.0003, "grad_norm": 0.14543126631900252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53967315924, "tokens_per_second": 148058.31468745435, "data_wait_s": 1.321800118603278e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368800, "loss": 0.5869172811508179, "lr": 0.0003, "grad_norm": 0.15332136855723302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53982090871, "tokens_per_second": 150386.07507671177, "data_wait_s": 1.2170999980298802e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 368900, "loss": 0.5812274729833007, "lr": 0.0003, "grad_norm": 0.1567969206131114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53996816231, "tokens_per_second": 150214.356637518, "data_wait_s": 1.3823999324813485e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a539b486b0d393521af452d49889f2db16c0d3ef27bc30902b3aa5041a022ec4
-size 294015
+oid sha256:c9cf47be5dded19e8eb06b28449902768bc506ace7dc7fb2567c1f80c35bc5a2
+size 296133