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": 75769.82437471514,
-      "required_pages_per_second_upper_estimate": 0.006567536339936117
+      "required_token_positions_per_second": 75079.16419408521,
+      "required_pages_per_second_upper_estimate": 0.006507671665941339
     },
     {
       "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": 9471.228046839393,
-      "required_pages_per_second_upper_estimate": 0.00377821447536277
+      "required_token_positions_per_second": 9384.895524260652,
+      "required_pages_per_second_upper_estimate": 0.0037437751413198707
     },
     {
       "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": 18942.456093678786,
-      "required_pages_per_second_upper_estimate": 0.004382003841431537
+      "required_token_positions_per_second": 18769.791048521303,
+      "required_pages_per_second_upper_estimate": 0.004342060822035361
     },
     {
       "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": 28413.684140518177,
-      "required_pages_per_second_upper_estimate": 0.0017302442890780008
+      "required_token_positions_per_second": 28154.686572781953,
+      "required_pages_per_second_upper_estimate": 0.001714472696058085
     },
     {
       "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": 28413.684140518177,
-      "required_pages_per_second_upper_estimate": 0.0021873640592488773
+      "required_token_positions_per_second": 28154.686572781953,
+      "required_pages_per_second_upper_estimate": 0.0021674257095333884
     },
     {
       "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": 28413.684140518177,
-      "required_pages_per_second_upper_estimate": 0.35280286254166626
+      "required_token_positions_per_second": 28154.686572781953,
+      "required_pages_per_second_upper_estimate": 0.34958697955948137
     }
   ],
   "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": 0.06874965441403447,
-    "producer_p95_seconds": 0.24045136799759348,
-    "gpu_update_seconds": 0.9730522752089615,
-    "producer_batches_per_second": 14.545527661530489,
-    "gpu_batches_per_second": 1.0276940154990661,
-    "producer_headroom": 14.153558785167128,
+    "producer_batch_seconds": 0.11454749972639888,
+    "producer_p95_seconds": 0.30944774000090547,
+    "gpu_update_seconds": 0.9820034731527865,
+    "producer_batches_per_second": 8.73000285810287,
+    "gpu_batches_per_second": 1.0183263372678657,
+    "producer_headroom": 8.57289312729077,
     "sustainable": true,
-    "buffer_seconds": 60.32924106295562,
+    "buffer_seconds": 60.88421533547276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390701.3278523637,
-    "upload_bytes_per_second": 5965009.527629528,
+    "generation_bytes_per_second": 347522.24006074615,
+    "upload_bytes_per_second": 5906802.196377309,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.83540989568885,
+    "estimated_full_shards_per_hour": 74.57018281332768,
     "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": 1.118036003999805,
-      "bytes": 11128347,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.0285996809998323,
       "bytes": 1163689,
@@ -562,6 +556,12 @@
       "bytes": 6281942,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0568676230031997,
+      "bytes": 6659990,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 100100, "loss": 0.6103905867785215, "lr": 0.0003, "grad_norm": 0.1358339739445815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14647761693, "tokens_per_second": 124416.37253416263, "data_wait_s": 1.0945001122308895e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 100200, "loss": 0.6084248721599579, "lr": 0.0003, "grad_norm": 0.13180838639884784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14662512988, "tokens_per_second": 150314.62526358146, "data_wait_s": 9.77099989540875e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 100300, "loss": 0.6339954510331154, "lr": 0.0003, "grad_norm": 0.13517477305933878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14677219940, "tokens_per_second": 141264.01372376605, "data_wait_s": 1.4471999747911468e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 100400, "loss": 0.6271757539361715, "lr": 0.0003, "grad_norm": 0.1342398332251596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14692038012, "tokens_per_second": 150552.59828249892, "data_wait_s": 1.7460002709412947e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc38b1824a603dada047bac441a5865eb15eb5941ee6b6724aaa1fa7d44526f9
-size 326568
+oid sha256:4c1e7fa1ab99aa9bf73e8a71ccd5979f33292704f038d847c70277395b1f42f7
+size 328686