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": 76376.25931618252,
-      "required_pages_per_second_upper_estimate": 0.0066201005836672025
+      "required_token_positions_per_second": 76003.94676086913,
+      "required_pages_per_second_upper_estimate": 0.0065878294749901825
     },
     {
       "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": 9547.032414522815,
-      "required_pages_per_second_upper_estimate": 0.0038084539710079844
+      "required_token_positions_per_second": 9500.493345108642,
+      "required_pages_per_second_upper_estimate": 0.0037898888403975753
     },
     {
       "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": 19094.06482904563,
-      "required_pages_per_second_upper_estimate": 0.004417075854136045
+      "required_token_positions_per_second": 19000.986690217283,
+      "required_pages_per_second_upper_estimate": 0.004395543864837402
     },
     {
       "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": 28641.09724356845,
-      "required_pages_per_second_upper_estimate": 0.0017440925539094218
+      "required_token_positions_per_second": 28501.480035325927,
+      "required_pages_per_second_upper_estimate": 0.0017355905984423321
     },
     {
       "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": 28641.09724356845,
-      "required_pages_per_second_upper_estimate": 0.0022048709494414476
+      "required_token_positions_per_second": 28501.480035325927,
+      "required_pages_per_second_upper_estimate": 0.0021941228302657695
     },
     {
       "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": 28641.09724356845,
-      "required_pages_per_second_upper_estimate": 0.35562657217885507
+      "required_token_positions_per_second": 28501.480035325927,
+      "required_pages_per_second_upper_estimate": 0.3538929937212204
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05389935951563274,
-    "producer_p95_seconds": 0.15546805500025584,
-    "gpu_update_seconds": 0.9653261453245666,
-    "producer_batches_per_second": 18.553096158962042,
-    "gpu_batches_per_second": 1.0359193158119375,
-    "producer_headroom": 17.90978879896685,
+    "producer_batch_seconds": 0.05427880309376221,
+    "producer_p95_seconds": 0.1454224730000533,
+    "gpu_update_seconds": 0.9700548871753998,
+    "producer_batches_per_second": 18.423398140754532,
+    "gpu_batches_per_second": 1.0308695035925175,
+    "producer_headroom": 17.87170740481711,
     "sustainable": true,
-    "buffer_seconds": 60.815547155447696,
+    "buffer_seconds": 60.143403004874784,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -21,3 +21,4 @@
 {"step": 117100, "loss": 0.6071493308991194, "lr": 0.0003, "grad_norm": 0.14773526847702778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17135712327, "tokens_per_second": 148386.77664299984, "data_wait_s": 1.4575000022887252e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 117200, "loss": 0.6492827702313662, "lr": 0.0003, "grad_norm": 0.15350586535446745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17150398058, "tokens_per_second": 151387.28465944977, "data_wait_s": 1.5145999896049034e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 117300, "loss": 0.6031660884618759, "lr": 0.0003, "grad_norm": 0.12929184356664308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17165055303, "tokens_per_second": 150776.66775610272, "data_wait_s": 1.257900021300884e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 117400, "loss": 0.5866196602582932, "lr": 0.0003, "grad_norm": 0.1448304762316454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17179656845, "tokens_per_second": 150238.43506845544, "data_wait_s": 1.3821000266034389e-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:f0a81bb7f493bdb1e3684bedaae4e796e633ec0dc40431a9c3df82d77c553b62
-size 48879
+oid sha256:4add9c4050f40ff18a80bf02a66ac4f2bc3be3b388caaad4de77cd276f37b492
+size 50997