Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77172.93108722073,
-      "required_pages_per_second_upper_estimate": 0.005339002438484515
+      "required_token_positions_per_second": 76165.18320351352,
+      "required_pages_per_second_upper_estimate": 0.0052692841015924054
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9646.616385902591,
-      "required_pages_per_second_upper_estimate": 0.004038836836520287
+      "required_token_positions_per_second": 9520.64790043919,
+      "required_pages_per_second_upper_estimate": 0.003986096462177864
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19293.232771805182,
-      "required_pages_per_second_upper_estimate": 0.004018271589603726
+      "required_token_positions_per_second": 19041.29580087838,
+      "required_pages_per_second_upper_estimate": 0.003965799762584388
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28939.849157707777,
-      "required_pages_per_second_upper_estimate": 0.00173489976841352
+      "required_token_positions_per_second": 28561.94370131757,
+      "required_pages_per_second_upper_estimate": 0.0017122449133311478
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28939.849157707777,
-      "required_pages_per_second_upper_estimate": 0.0022499459790680677
+      "required_token_positions_per_second": 28561.94370131757,
+      "required_pages_per_second_upper_estimate": 0.002220565492064162
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28939.849157707777,
-      "required_pages_per_second_upper_estimate": 0.3712093118060027
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28561.94370131757,
+      "required_pages_per_second_upper_estimate": 0.36636194637469466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.24199939645314572,
-    "producer_p95_seconds": 0.17852025000047433,
-    "gpu_update_seconds": 0.9553608883492156,
-    "producer_batches_per_second": 4.13224171074994,
-    "gpu_batches_per_second": 1.046724868262,
-    "producer_headroom": 3.9477821116557457,
+    "producer_batch_seconds": 0.17652099367968077,
+    "producer_p95_seconds": 0.20069275100013328,
+    "gpu_update_seconds": 0.968001347846806,
+    "producer_batches_per_second": 5.665048554024254,
+    "gpu_batches_per_second": 1.033056412808072,
+    "producer_headroom": 5.483774635913077,
     "sustainable": true,
-    "buffer_seconds": 60.18773596600058,
+    "buffer_seconds": 60.016083566501976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 155,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 592445.3160897726,
-    "upload_bytes_per_second": 2726643.835663862,
+    "generation_bytes_per_second": 582705.1219094099,
+    "upload_bytes_per_second": 2783607.079390321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.562486705874846,
+    "estimated_full_shards_per_hour": 62.51747724037992,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 6215302,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2488605079997797,
+      "bytes": 6396111,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -68,3 +68,4 @@
 {"step": 691800, "loss": 0.5641340017318726, "lr": 8e-05, "grad_norm": 0.19846600156899089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101261156602, "tokens_per_second": 150403.5372260531, "data_wait_s": 1.6255999980785418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691900, "loss": 0.5252321362495422, "lr": 8e-05, "grad_norm": 0.19011193435292703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101275755351, "tokens_per_second": 152201.3285828158, "data_wait_s": 1.1328999789839145e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692000, "loss": 0.5682318806648254, "lr": 8e-05, "grad_norm": 0.19148309451636908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101290222071, "tokens_per_second": 150757.88974356095, "data_wait_s": 1.6043999494286254e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 692100, "loss": 0.5424365401268005, "lr": 8e-05, "grad_norm": 0.2043193890694023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101304951523, "tokens_per_second": 153630.61658268468, "data_wait_s": 1.5503999748034403e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba4d1c8a76830a2800f76c3579c567e0b0d5f979058aec38d04a5bb471732faa
-size 149208
+oid sha256:f1bb3c3a2d82a0928c7175122b374e06cf0ac99bb8ab399adfc42d546b1467ac
+size 151326