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": 74303.1604801283,
-      "required_pages_per_second_upper_estimate": 0.006440409630778986
+      "required_token_positions_per_second": 74077.04113593441,
+      "required_pages_per_second_upper_estimate": 0.0064208101791184986
     },
     {
       "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": 9287.895060016037,
-      "required_pages_per_second_upper_estimate": 0.003705080205846512
+      "required_token_positions_per_second": 9259.630141991802,
+      "required_pages_per_second_upper_estimate": 0.0036938049074484607
     },
     {
       "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": 18575.790120032074,
-      "required_pages_per_second_upper_estimate": 0.0042971821215291054
+      "required_token_positions_per_second": 18519.260283983604,
+      "required_pages_per_second_upper_estimate": 0.004284104938850427
     },
     {
       "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": 27863.685180048113,
-      "required_pages_per_second_upper_estimate": 0.0016967522380069069
+      "required_token_positions_per_second": 27778.89042597541,
+      "required_pages_per_second_upper_estimate": 0.0016915886823675715
     },
     {
       "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": 27863.685180048113,
-      "required_pages_per_second_upper_estimate": 0.002145023616777325
+      "required_token_positions_per_second": 27778.89042597541,
+      "required_pages_per_second_upper_estimate": 0.002138495881881906
     },
     {
       "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": 27863.685180048113,
-      "required_pages_per_second_upper_estimate": 0.3459737161807382
+      "required_token_positions_per_second": 27778.89042597541,
+      "required_pages_per_second_upper_estimate": 0.3449208491249414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10627422174994194,
-    "producer_p95_seconds": 0.3385915590006334,
-    "gpu_update_seconds": 0.992259273005189,
-    "producer_batches_per_second": 9.409619600442252,
-    "gpu_batches_per_second": 1.0078011132829903,
-    "producer_headroom": 9.336782303990207,
+    "producer_batch_seconds": 0.09899960610144376,
+    "producer_p95_seconds": 0.3328307859992492,
+    "gpu_update_seconds": 0.995288133400281,
+    "producer_batches_per_second": 10.101050290799254,
+    "gpu_batches_per_second": 1.0047341733932078,
+    "producer_headroom": 10.053455489311954,
     "sustainable": true,
-    "buffer_seconds": 60.52781565331653,
+    "buffer_seconds": 60.71257613741714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385591.8119941518,
-    "upload_bytes_per_second": 2866688.096561006,
+    "generation_bytes_per_second": 420089.4376619857,
+    "upload_bytes_per_second": 2834547.5719834934,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.73902673595825,
+    "estimated_full_shards_per_hour": 90.14141414065054,
     "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": 1.0683662000010372,
-      "bytes": 6645182,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0938033080019522,
       "bytes": 6719573,
@@ -562,6 +556,12 @@
       "bytes": 5472606,
       "pages": 16,
       "shards": 2
+    },
+    {
+      "seconds": 0.7334500290016877,
+      "bytes": 861667,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -250,3 +250,4 @@
 {"step": 220000, "loss": 0.6188193764537573, "lr": 0.0003, "grad_norm": 0.1620542855839367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32201121996, "tokens_per_second": 149783.29342078025, "data_wait_s": 1.0130999726243317e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 220100, "loss": 0.6336135417222977, "lr": 0.0003, "grad_norm": 0.15108254080626116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32215851555, "tokens_per_second": 140281.62330591143, "data_wait_s": 1.1740998161258176e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 220200, "loss": 0.619529539719224, "lr": 0.0003, "grad_norm": 0.1527795924355659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32230628597, "tokens_per_second": 150216.34231026022, "data_wait_s": 9.638999472372234e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 220300, "loss": 0.5912884678691626, "lr": 0.0003, "grad_norm": 0.13969897268591489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32245296649, "tokens_per_second": 149061.06156735765, "data_wait_s": 8.604998583905399e-06, "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 sha256:7a3f072720228b59710ceac3fa1f0c670f5493684c0ea2251860ddb0d79a3558
-size 537816
+oid sha256:4e9f9a087cf01e3362af8e9a82bf8858930e007208e604cc3d400ab91dbc69a5
+size 539934