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": 75863.0953044941,
-      "required_pages_per_second_upper_estimate": 0.006575620827736346
+      "required_token_positions_per_second": 75945.12786852336,
+      "required_pages_per_second_upper_estimate": 0.00658273120247677
     },
     {
       "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": 9482.886913061762,
-      "required_pages_per_second_upper_estimate": 0.0037828653714144576
+      "required_token_positions_per_second": 9493.14098356542,
+      "required_pages_per_second_upper_estimate": 0.0037869558734503826
     },
     {
       "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": 18965.773826123524,
-      "required_pages_per_second_upper_estimate": 0.004387397988454553
+      "required_token_positions_per_second": 18986.28196713084,
+      "required_pages_per_second_upper_estimate": 0.004392142185945641
     },
     {
       "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": 28448.660739185285,
-      "required_pages_per_second_upper_estimate": 0.0017323741804287968
+      "required_token_positions_per_second": 28479.422950696255,
+      "required_pages_per_second_upper_estimate": 0.0017342474377129644
     },
     {
       "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": 28448.660739185285,
-      "required_pages_per_second_upper_estimate": 0.0021900566546356934
+      "required_token_positions_per_second": 28479.422950696255,
+      "required_pages_per_second_upper_estimate": 0.002192424814833059
     },
     {
       "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": 28448.660739185285,
-      "required_pages_per_second_upper_estimate": 0.3532371548379662
+      "required_token_positions_per_second": 28479.422950696255,
+      "required_pages_per_second_upper_estimate": 0.3536191185504334
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.03927175721871379,
-    "producer_p95_seconds": 0.13259731200014357,
-    "gpu_update_seconds": 0.9718559426566449,
-    "producer_batches_per_second": 25.463591925127297,
-    "gpu_batches_per_second": 1.0289590834485418,
-    "producer_headroom": 24.746943133818718,
+    "producer_batch_seconds": 0.040468276663943925,
+    "producer_p95_seconds": 0.1352899539997452,
+    "gpu_update_seconds": 0.9708061869042915,
+    "producer_batches_per_second": 24.710713735210064,
+    "gpu_batches_per_second": 1.0300717213070116,
+    "producer_headroom": 23.989313776962785,
     "sustainable": true,
-    "buffer_seconds": 60.25506844471198,
+    "buffer_seconds": 60.18998358806608,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -28,3 +28,4 @@
 {"step": 87800, "loss": 0.6800566706806421, "lr": 0.0003, "grad_norm": 0.13805183768005552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12848973134, "tokens_per_second": 151097.18100490732, "data_wait_s": 8.82499989529606e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 87900, "loss": 0.6352664791047573, "lr": 0.0003, "grad_norm": 0.13876333872899696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12863805001, "tokens_per_second": 151854.48770500845, "data_wait_s": 9.638999472372234e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 88000, "loss": 0.5933379828929901, "lr": 0.0003, "grad_norm": 0.13084436984565131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12878446667, "tokens_per_second": 149637.3131397905, "data_wait_s": 1.1097999959019944e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 88100, "loss": 0.6502611301839352, "lr": 0.0003, "grad_norm": 0.15064749037323957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12893146289, "tokens_per_second": 150080.77260234006, "data_wait_s": 1.1208001524209976e-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 sha256:8efed257c085c39094c0e3298872a398c265ff7840284833d381d6fea36d4dd8
-size 63705
+oid sha256:f9000820f26a27e214e19a61b309b4ff6a06f8c3f3b7ccb80a7aa6c6d4ef1733
+size 65823