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": 75207.32936979407,
-      "required_pages_per_second_upper_estimate": 0.006518780698540115
+      "required_token_positions_per_second": 75358.5207329697,
+      "required_pages_per_second_upper_estimate": 0.006531885582709172
     },
     {
       "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": 9400.916171224259,
-      "required_pages_per_second_upper_estimate": 0.003750166016923671
+      "required_token_positions_per_second": 9419.815091621213,
+      "required_pages_per_second_upper_estimate": 0.0037577050788340565
     },
     {
       "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": 18801.832342448517,
-      "required_pages_per_second_upper_estimate": 0.004349473011477908
+      "required_token_positions_per_second": 18839.630183242425,
+      "required_pages_per_second_upper_estimate": 0.004358216876726301
     },
     {
       "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": 28202.748513672777,
-      "required_pages_per_second_upper_estimate": 0.0017173994160968203
+      "required_token_positions_per_second": 28259.445274863636,
+      "required_pages_per_second_upper_estimate": 0.0017208519513884268
     },
     {
       "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": 28202.748513672777,
-      "required_pages_per_second_upper_estimate": 0.0021711256507871315
+      "required_token_positions_per_second": 28259.445274863636,
+      "required_pages_per_second_upper_estimate": 0.0021754903244102838
     },
     {
       "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": 28202.748513672777,
-      "required_pages_per_second_upper_estimate": 0.3501837480123766
+      "required_token_positions_per_second": 28259.445274863636,
+      "required_pages_per_second_upper_estimate": 0.3508877320345138
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1083144178518296,
-    "producer_p95_seconds": 0.32796919599786634,
-    "gpu_update_seconds": 0.9803299840296121,
-    "producer_batches_per_second": 9.232381245569409,
-    "gpu_batches_per_second": 1.0200646887179101,
-    "producer_headroom": 9.050780159024349,
+    "producer_batch_seconds": 0.3797645783203336,
+    "producer_p95_seconds": 0.32158002500000293,
+    "gpu_update_seconds": 0.9783631536671561,
+    "producer_batches_per_second": 2.6332103020848203,
+    "gpu_batches_per_second": 1.0221153528234823,
+    "producer_headroom": 2.576235935416549,
     "sustainable": true,
-    "buffer_seconds": 60.78045900983595,
+    "buffer_seconds": 60.658515527363676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343888.7414981482,
-    "upload_bytes_per_second": 5056099.634686686,
+    "generation_bytes_per_second": 322001.1410820949,
+    "upload_bytes_per_second": 4905317.934016314,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.79051860531172,
+    "estimated_full_shards_per_hour": 69.09394907328735,
     "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": 1.406141793999268,
-      "bytes": 13345421,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0539056280031218,
       "bytes": 6000362,
@@ -562,6 +556,12 @@
       "bytes": 6738522,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 2.4403519309998956,
+      "bytes": 5986840,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 261900, "loss": 0.585339205339551, "lr": 0.0003, "grad_norm": 0.1740283848217525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38335035150, "tokens_per_second": 148937.061681626, "data_wait_s": 1.7112000932684168e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262000, "loss": 0.5359991174191236, "lr": 0.0003, "grad_norm": 0.15083570431940835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38349519787, "tokens_per_second": 147611.5745483135, "data_wait_s": 1.280799915548414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262100, "loss": 0.5910497801378369, "lr": 0.0003, "grad_norm": 0.1564822698806592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38364018932, "tokens_per_second": 147168.54877804554, "data_wait_s": 1.5912002709228545e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 262200, "loss": 0.6033521145582199, "lr": 0.0003, "grad_norm": 0.14459503681200822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38378590332, "tokens_per_second": 147539.9967850133, "data_wait_s": 1.3729997590417042e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:297e612e145028c994d2567373534790b00899d4a0b01857a115f39611d2e64d
-size 533580
+oid sha256:de5e0f9daaf23dd83bf216e18c116f8aee050c59f0b9bbfb46353766a778f650
+size 535698