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": 75198.09000862221,
-      "required_pages_per_second_upper_estimate": 0.006517979854130685
+      "required_token_positions_per_second": 76030.18880174283,
+      "required_pages_per_second_upper_estimate": 0.0065901040685832695
     },
     {
       "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": 9399.761251077776,
-      "required_pages_per_second_upper_estimate": 0.003749705302009644
+      "required_token_positions_per_second": 9503.773600217854,
+      "required_pages_per_second_upper_estimate": 0.0037911973832048243
     },
     {
       "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": 18799.522502155553,
-      "required_pages_per_second_upper_estimate": 0.004348938670577931
+      "required_token_positions_per_second": 19007.547200435707,
+      "required_pages_per_second_upper_estimate": 0.004397061523415398
     },
     {
       "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": 28199.28375323333,
-      "required_pages_per_second_upper_estimate": 0.001717188430364251
+      "required_token_positions_per_second": 28511.32080065356,
+      "required_pages_per_second_upper_estimate": 0.0017361898494202795
     },
     {
       "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": 28199.28375323333,
-      "required_pages_per_second_upper_estimate": 0.002170858923937432
+      "required_token_positions_per_second": 28511.32080065356,
+      "required_pages_per_second_upper_estimate": 0.0021948803996216727
     },
     {
       "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": 28199.28375323333,
-      "required_pages_per_second_upper_estimate": 0.35014072728352597
+      "required_token_positions_per_second": 28511.32080065356,
+      "required_pages_per_second_upper_estimate": 0.3540151830916667
     }
   ],
   "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": 0.14977265961022113,
-    "producer_p95_seconds": 0.36263739800051553,
-    "gpu_update_seconds": 0.9804504342004744,
-    "producer_batches_per_second": 6.676786020909758,
-    "gpu_batches_per_second": 1.0199393718617378,
-    "producer_headroom": 6.54625775326463,
+    "producer_batch_seconds": 0.09402382284417854,
+    "producer_p95_seconds": 0.3534700279997196,
+    "gpu_update_seconds": 0.96972006990873,
+    "producer_batches_per_second": 10.635602443619582,
+    "gpu_batches_per_second": 1.0312254340514164,
+    "producer_headroom": 10.313557145148241,
     "sustainable": true,
-    "buffer_seconds": 60.78792692042941,
+    "buffer_seconds": 60.12264433434126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407072.2213428866,
-    "upload_bytes_per_second": 5210661.099102002,
+    "generation_bytes_per_second": 394297.48852221004,
+    "upload_bytes_per_second": 5056106.653577283,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.348222543859,
+    "estimated_full_shards_per_hour": 84.60706226110197,
     "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": 1.163448106002761,
-      "bytes": 13378237,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 0.9504098300021724,
       "bytes": 828896,
@@ -562,6 +556,12 @@
       "bytes": 7821336,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2159917039971333,
+      "bytes": 870322,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 192300, "loss": 0.5848005227744579, "lr": 0.0003, "grad_norm": 0.15712572571967284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28145089960, "tokens_per_second": 149547.05379373714, "data_wait_s": 1.1344003723934293e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 192400, "loss": 0.5970455035567284, "lr": 0.0003, "grad_norm": 0.15022067770491687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28159706591, "tokens_per_second": 149849.2442474602, "data_wait_s": 1.0934003512375057e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 192500, "loss": 0.6084013562649488, "lr": 0.0003, "grad_norm": 0.17986228027761936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28174428544, "tokens_per_second": 151836.07050376487, "data_wait_s": 1.0881994967348874e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 192600, "loss": 0.5691842325031757, "lr": 0.0003, "grad_norm": 0.12877266319356181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28189094236, "tokens_per_second": 150091.31227574433, "data_wait_s": 9.251001756638288e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6651462a495f5ab5bedd7a921c06185a3d99717ea0bcac757cb3de94d6885b23
-size 799896
+oid sha256:28842941b79700dea27103cc59d7d55df4fa97ade253b2e708bd9eb8e4c7181b
+size 802014