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": 75812.52095491662,
-      "required_pages_per_second_upper_estimate": 0.006571237171294488
+      "required_token_positions_per_second": 75986.72892047396,
+      "required_pages_per_second_upper_estimate": 0.006586337076222947
     },
     {
       "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": 9476.565119364577,
-      "required_pages_per_second_upper_estimate": 0.0037803435133894115
+      "required_token_positions_per_second": 9498.341115059246,
+      "required_pages_per_second_upper_estimate": 0.0037890302836521643
     },
     {
       "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": 18953.130238729154,
-      "required_pages_per_second_upper_estimate": 0.0043844731170304034
+      "required_token_positions_per_second": 18996.68223011849,
+      "required_pages_per_second_upper_estimate": 0.004394548103749457
     },
     {
       "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": 28429.695358093733,
-      "required_pages_per_second_upper_estimate": 0.001731219288223986
+      "required_token_positions_per_second": 28495.023345177735,
+      "required_pages_per_second_upper_estimate": 0.0017351974198879423
     },
     {
       "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": 28429.695358093733,
-      "required_pages_per_second_upper_estimate": 0.0021885966470997307
+      "required_token_positions_per_second": 28495.023345177735,
+      "required_pages_per_second_upper_estimate": 0.0021936257763848998
     },
     {
       "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": 28429.695358093733,
-      "required_pages_per_second_upper_estimate": 0.35300166827785656
+      "required_token_positions_per_second": 28495.023345177735,
+      "required_pages_per_second_upper_estimate": 0.3538128232387317
     }
   ],
   "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": 0.153072793062222,
-    "producer_p95_seconds": 0.38399713700346183,
-    "gpu_update_seconds": 0.9725042654081346,
-    "producer_batches_per_second": 6.532839572565411,
-    "gpu_batches_per_second": 1.0282731249310522,
-    "producer_headroom": 6.353214349546918,
+    "producer_batch_seconds": 0.34786596828917027,
+    "producer_p95_seconds": 0.3888129650003975,
+    "gpu_update_seconds": 0.9702746920078913,
+    "producer_batches_per_second": 2.8746703936521056,
+    "gpu_batches_per_second": 1.030635971686116,
+    "producer_headroom": 2.7892199308250003,
     "sustainable": true,
-    "buffer_seconds": 60.29526445530435,
+    "buffer_seconds": 60.15703090448926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402890.5873529188,
-    "upload_bytes_per_second": 5372402.5939944815,
+    "generation_bytes_per_second": 382708.4029622491,
+    "upload_bytes_per_second": 5479111.05094803,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.45094123306916,
+    "estimated_full_shards_per_hour": 82.12031428003888,
     "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": 4.007056702997943,
-      "bytes": 13411793,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.3467935920016316,
       "bytes": 10736953,
@@ -562,6 +556,12 @@
       "bytes": 6155396,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1965257189949625,
+      "bytes": 6621394,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -338,3 +338,4 @@
 {"step": 148800, "loss": 0.5818050615489483, "lr": 0.0003, "grad_norm": 0.13499724230614996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21776061607, "tokens_per_second": 149635.5790251819, "data_wait_s": 1.7676997231319547e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 148900, "loss": 0.6206430681049824, "lr": 0.0003, "grad_norm": 0.13992829151953254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21790770150, "tokens_per_second": 150523.07218353453, "data_wait_s": 1.3739001587964594e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149000, "loss": 0.5806967727839947, "lr": 0.0003, "grad_norm": 0.13410837133544523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21805523148, "tokens_per_second": 150535.1928402669, "data_wait_s": 1.2776996300090104e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 149100, "loss": 0.5811381638050079, "lr": 0.0003, "grad_norm": 0.13801216699145583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21820213300, "tokens_per_second": 149975.18766694973, "data_wait_s": 1.0394003766123205e-05, "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 sha256:76c012da262b42bf0c2ca57dec1e90b0e48dbb35684d39e8734458ea333500bb
-size 724983
+oid sha256:636c1259ce98cae8faa46d77f69126ddfdb5d9f76d4bc9c9a7ab3405c7ae67d2
+size 727101