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": 75815.5412637281,
-      "required_pages_per_second_upper_estimate": 0.006571498964007364
+      "required_token_positions_per_second": 75476.74847850065,
+      "required_pages_per_second_upper_estimate": 0.006542133264046313
     },
     {
       "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": 9476.942657966012,
-      "required_pages_per_second_upper_estimate": 0.0037804941191822293
+      "required_token_positions_per_second": 9434.593559812582,
+      "required_pages_per_second_upper_estimate": 0.0037636004307533832
     },
     {
       "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": 18953.885315932024,
-      "required_pages_per_second_upper_estimate": 0.004384647790852382
+      "required_token_positions_per_second": 18869.187119625163,
+      "required_pages_per_second_upper_estimate": 0.004365054353774133
     },
     {
       "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": 28430.827973898035,
-      "required_pages_per_second_upper_estimate": 0.001731288258584092
+      "required_token_positions_per_second": 28303.780679437747,
+      "required_pages_per_second_upper_estimate": 0.0017235517449170996
     },
     {
       "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": 28430.827973898035,
-      "required_pages_per_second_upper_estimate": 0.002188683838999627
+      "required_token_positions_per_second": 28303.780679437747,
+      "required_pages_per_second_upper_estimate": 0.0021789033865826537
     },
     {
       "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": 28430.827973898035,
-      "required_pages_per_second_upper_estimate": 0.35301573157552474
+      "required_token_positions_per_second": 28303.780679437747,
+      "required_pages_per_second_upper_estimate": 0.3514382293782702
     }
   ],
   "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": 0.09079362663260326,
-    "producer_p95_seconds": 0.24800184000196168,
-    "gpu_update_seconds": 0.97246552317728,
-    "producer_batches_per_second": 11.01398894491244,
-    "gpu_batches_per_second": 1.0283140904911037,
-    "producer_headroom": 10.710724521583053,
+    "producer_batch_seconds": 0.255691637929516,
+    "producer_p95_seconds": 0.3173454949974257,
+    "gpu_update_seconds": 0.9768306330922724,
+    "producer_batches_per_second": 3.9109609062602986,
+    "gpu_batches_per_second": 1.0237189192504972,
+    "producer_headroom": 3.8203464180613746,
     "sustainable": true,
-    "buffer_seconds": 60.292862436991356,
+    "buffer_seconds": 60.563499251720884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 551678.658731242,
-    "upload_bytes_per_second": 4030880.544225077,
+    "generation_bytes_per_second": 517443.7798348389,
+    "upload_bytes_per_second": 4044581.1538177794,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.37739774182266,
+    "estimated_full_shards_per_hour": 111.03138967784763,
     "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": 1.2985985390005226,
-      "bytes": 6343213,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.028006202999677,
       "bytes": 2590442,
@@ -562,6 +556,12 @@
       "bytes": 6210420,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0871770170015225,
+      "bytes": 6926231,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 135700, "loss": 0.5719640366733074, "lr": 0.0003, "grad_norm": 0.1361328808762725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19859178115, "tokens_per_second": 149631.42355862632, "data_wait_s": 1.4367000403581187e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 135800, "loss": 0.6057610101997852, "lr": 0.0003, "grad_norm": 0.1535781581093354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19873871396, "tokens_per_second": 150192.94375282497, "data_wait_s": 1.3393000699579716e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 135900, "loss": 0.6181537136435509, "lr": 0.0003, "grad_norm": 0.15657916734791175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19888509149, "tokens_per_second": 148575.27425753482, "data_wait_s": 1.375000283587724e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 136000, "loss": 0.6349450200796127, "lr": 0.0003, "grad_norm": 0.14239278688334917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19903261936, "tokens_per_second": 150953.43991890686, "data_wait_s": 9.090999810723588e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a95bc6b235ff156c6840f56a4989554b7b7ac0bb7689a42b18afe084dfee8a9b
-size 445959
+oid sha256:26a9c9a2c6647c44e65ff0345713cc3b0d18c2135c601048d5ac4cbedfd51141
+size 448077