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": 74699.91833092092,
-      "required_pages_per_second_upper_estimate": 0.0064747995957121105
+      "required_token_positions_per_second": 74739.37631603684,
+      "required_pages_per_second_upper_estimate": 0.00647821971385661
     },
     {
       "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": 9337.489791365115,
-      "required_pages_per_second_upper_estimate": 0.003724864285688972
+      "required_token_positions_per_second": 9342.422039504605,
+      "required_pages_per_second_upper_estimate": 0.003726831833215496
     },
     {
       "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": 18674.97958273023,
-      "required_pages_per_second_upper_estimate": 0.004320127858049402
+      "required_token_positions_per_second": 18684.84407900921,
+      "required_pages_per_second_upper_estimate": 0.00432240983565434
     },
     {
       "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": 28012.469374095348,
-      "required_pages_per_second_upper_estimate": 0.001705812414814046
+      "required_token_positions_per_second": 28027.266118513813,
+      "required_pages_per_second_upper_estimate": 0.0017067134589166125
     },
     {
       "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": 28012.469374095348,
-      "required_pages_per_second_upper_estimate": 0.002156477435896091
+      "required_token_positions_per_second": 28027.266118513813,
+      "required_pages_per_second_upper_estimate": 0.0021576165302414343
     },
     {
       "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": 28012.469374095348,
-      "required_pages_per_second_upper_estimate": 0.3478211179221395
+      "required_token_positions_per_second": 28027.266118513813,
+      "required_pages_per_second_upper_estimate": 0.3480048439662989
     }
   ],
   "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": 0.05888063563281776,
-    "producer_p95_seconds": 0.20180956800140848,
-    "gpu_update_seconds": 0.9869890308766427,
-    "producer_batches_per_second": 16.98351230846155,
-    "gpu_batches_per_second": 1.0131824860422218,
-    "producer_headroom": 16.762540354209996,
+    "producer_batch_seconds": 0.13463949000774278,
+    "producer_p95_seconds": 0.22402411199982453,
+    "gpu_update_seconds": 0.9864679588472853,
+    "producer_batches_per_second": 7.427241442629444,
+    "gpu_batches_per_second": 1.0137176692170795,
+    "producer_headroom": 7.326735705776634,
     "sustainable": true,
-    "buffer_seconds": 60.206330883475204,
+    "buffer_seconds": 60.174545489684405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349718.9336879294,
-    "upload_bytes_per_second": 5260384.973756027,
+    "generation_bytes_per_second": 354985.785571054,
+    "upload_bytes_per_second": 5275946.098758672,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.04154212931073,
+    "estimated_full_shards_per_hour": 76.17168593739237,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 1209843,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.586552576000031,
+      "bytes": 9572867,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -104,3 +104,4 @@
 {"step": 205400, "loss": 0.6149740368127823, "lr": 0.0003, "grad_norm": 0.13280079023353883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30062883332, "tokens_per_second": 147862.43994892322, "data_wait_s": 2.2259999241214246e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205500, "loss": 0.6220797058194876, "lr": 0.0003, "grad_norm": 0.14416894672017122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30077606188, "tokens_per_second": 148590.61643834022, "data_wait_s": 1.797600089048501e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205600, "loss": 0.5864623375236988, "lr": 0.0003, "grad_norm": 0.17855335037674847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30092330583, "tokens_per_second": 148678.9822647812, "data_wait_s": 1.805900137696881e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 205700, "loss": 0.6235309913754463, "lr": 0.0003, "grad_norm": 0.149813195949142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30107028663, "tokens_per_second": 147879.35779941638, "data_wait_s": 1.8307000573258847e-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 sha256:163321637941f018e7c6c7d68b170a099aaec7880795c299616a68a653fd5491
-size 226239
+oid sha256:f4c4d65cde32ae44cb565cbcc2a555d9bd35ada5c898fdb68ba1d4e54a1b47b2
+size 228357