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": 75858.14057968512,
-      "required_pages_per_second_upper_estimate": 0.00657519136474754
+      "required_token_positions_per_second": 75754.06320532336,
+      "required_pages_per_second_upper_estimate": 0.006566170201719556
     },
     {
       "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": 9482.26757246064,
-      "required_pages_per_second_upper_estimate": 0.0037826183071887026
+      "required_token_positions_per_second": 9469.25790066542,
+      "required_pages_per_second_upper_estimate": 0.0037774285545976623
     },
     {
       "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": 18964.53514492128,
-      "required_pages_per_second_upper_estimate": 0.0043871114413584576
+      "required_token_positions_per_second": 18938.51580133084,
+      "required_pages_per_second_upper_estimate": 0.004381092324143616
     },
     {
       "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": 28446.802717381917,
-      "required_pages_per_second_upper_estimate": 0.0017322610366492586
+      "required_token_positions_per_second": 28407.77370199626,
+      "required_pages_per_second_upper_estimate": 0.0017298843743816905
     },
     {
       "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": 28446.802717381917,
-      "required_pages_per_second_upper_estimate": 0.0021899136189739367
+      "required_token_positions_per_second": 28407.77370199626,
+      "required_pages_per_second_upper_estimate": 0.002186909057330319
     },
     {
       "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": 28446.802717381917,
-      "required_pages_per_second_upper_estimate": 0.35321408442556734
+      "required_token_positions_per_second": 28407.77370199626,
+      "required_pages_per_second_upper_estimate": 0.3527294746761893
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08437987716408202,
-    "producer_p95_seconds": 0.19607269200059818,
-    "gpu_update_seconds": 0.9719194200726881,
-    "producer_batches_per_second": 11.851166813806053,
-    "gpu_batches_per_second": 1.0288918806923437,
-    "producer_headroom": 11.518379176859066,
+    "producer_batch_seconds": 0.22961833171095236,
+    "producer_p95_seconds": 0.2572298109989788,
+    "gpu_update_seconds": 0.9732547256266383,
+    "producer_batches_per_second": 4.355052980956319,
+    "gpu_batches_per_second": 1.0274802409576194,
+    "producer_headroom": 4.238575894070116,
     "sustainable": true,
-    "buffer_seconds": 60.25900404450666,
+    "buffer_seconds": 60.34179298885157,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 511845.4954419861,
-    "upload_bytes_per_second": 5859617.074481066,
+    "generation_bytes_per_second": 484581.63552683487,
+    "upload_bytes_per_second": 5928758.260247174,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.83012816853224,
+    "estimated_full_shards_per_hour": 103.97993850091729,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 2646898,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0982734540011734,
+      "bytes": 10287883,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 95700, "loss": 0.5961932484060526, "lr": 0.0003, "grad_norm": 0.13956989449913731, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14004427927, "tokens_per_second": 148572.77696347405, "data_wait_s": 1.5059002180350944e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95800, "loss": 0.6572305075824261, "lr": 0.0003, "grad_norm": 0.14091792533270156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14019097985, "tokens_per_second": 150068.8900955594, "data_wait_s": 8.089999028015882e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 95900, "loss": 0.6055783797055483, "lr": 0.0003, "grad_norm": 0.15178553308913817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14033585492, "tokens_per_second": 147557.5283446402, "data_wait_s": 9.029998182086274e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 96000, "loss": 0.6212404612451792, "lr": 0.0003, "grad_norm": 0.1366567958379472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14048197658, "tokens_per_second": 149438.04337328838, "data_wait_s": 1.1753996659535915e-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 sha256:5f34887150e788f6befc3f2d848a84254d84cd98fe78c03eec3775f08489cfe2
-size 232593
+oid sha256:60173b826a61bc0a36262b9ca55b578f312b08f7796f041a0effc713558c7a90
+size 234711