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": 76538.4956437177,
-      "required_pages_per_second_upper_estimate": 0.005295110722409932
+      "required_token_positions_per_second": 77081.5214767311,
+      "required_pages_per_second_upper_estimate": 0.005332678509531324
     },
     {
       "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": 9567.311955464713,
-      "required_pages_per_second_upper_estimate": 0.004005633727560773
+      "required_token_positions_per_second": 9635.190184591387,
+      "required_pages_per_second_upper_estimate": 0.004034052924637503
     },
     {
       "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": 19134.623910929426,
-      "required_pages_per_second_upper_estimate": 0.003985237546680249
+      "required_token_positions_per_second": 19270.380369182774,
+      "required_pages_per_second_upper_estimate": 0.0040135120367958636
     },
     {
       "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": 28701.93586639414,
-      "required_pages_per_second_upper_estimate": 0.001720637229871829
+      "required_token_positions_per_second": 28905.570553774163,
+      "required_pages_per_second_upper_estimate": 0.0017328448184480974
     },
     {
       "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": 28701.93586639414,
-      "required_pages_per_second_upper_estimate": 0.002231449267138407
+      "required_token_positions_per_second": 28905.570553774163,
+      "required_pages_per_second_upper_estimate": 0.0022472809683878737
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28701.93586639414,
-      "required_pages_per_second_upper_estimate": 0.368157615556421
+      "required_token_positions_per_second": 28905.570553774163,
+      "required_pages_per_second_upper_estimate": 0.37076962268023966
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12152001985907646,
-    "producer_p95_seconds": 0.23447467399819288,
-    "gpu_update_seconds": 0.9632799727760473,
-    "producer_batches_per_second": 8.229096746031423,
-    "gpu_batches_per_second": 1.0381197868342786,
-    "producer_headroom": 7.926924089488608,
+    "producer_batch_seconds": 0.10733673492950402,
+    "producer_p95_seconds": 0.2267190339989611,
+    "gpu_update_seconds": 0.9564938338983949,
+    "producer_batches_per_second": 9.316474929638712,
+    "gpu_batches_per_second": 1.045485046071114,
+    "producer_headroom": 8.911150823868411,
     "sustainable": true,
-    "buffer_seconds": 60.686638284890975,
+    "buffer_seconds": 60.25911153559888,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348996.7839421241,
-    "upload_bytes_per_second": 5318701.754225176,
+    "generation_bytes_per_second": 348826.37383720104,
+    "upload_bytes_per_second": 5454933.030257024,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.88658560464661,
+    "estimated_full_shards_per_hour": 74.85001956307434,
     "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": 1.7196641009995801,
-      "bytes": 2291853,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0487870670003758,
       "bytes": 11950097,
@@ -562,6 +556,12 @@
       "bytes": 867774,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1689961140000378,
+      "bytes": 10341916,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 575200, "loss": 0.5646545886993408, "lr": 0.0003, "grad_norm": 0.14941126424010304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84192175237, "tokens_per_second": 152914.95770937504, "data_wait_s": 1.4995002857176587e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575300, "loss": 0.5778487920761108, "lr": 0.0003, "grad_norm": 0.1714084622270154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84206682974, "tokens_per_second": 151256.1242798681, "data_wait_s": 1.2793003406841308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575400, "loss": 0.5910930037498474, "lr": 0.0003, "grad_norm": 0.17306087163264713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84221195839, "tokens_per_second": 151026.81385286932, "data_wait_s": 1.5139998140512034e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 575500, "loss": 0.561769962310791, "lr": 0.0003, "grad_norm": 0.15479443685832425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84235675989, "tokens_per_second": 151186.59097593135, "data_wait_s": 1.3323999155545607e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39c05e92f4b4c72c0263f01bcdc6af4a9f8b7b370fd16a27f57d66124b895d00
-size 390891
+oid sha256:1d4eae14b9c2297d3fad1c49c626f4c1cc4270ea92c0597ba4503c36bb40e2f1
+size 393009