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": 76165.18320351352,
-      "required_pages_per_second_upper_estimate": 0.0052692841015924054
+      "required_token_positions_per_second": 76855.26663327214,
+      "required_pages_per_second_upper_estimate": 0.005317025674477268
     },
     {
       "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": 9520.64790043919,
-      "required_pages_per_second_upper_estimate": 0.003986096462177864
+      "required_token_positions_per_second": 9606.908329159018,
+      "required_pages_per_second_upper_estimate": 0.004022211902360269
     },
     {
       "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": 19041.29580087838,
-      "required_pages_per_second_upper_estimate": 0.003965799762584388
+      "required_token_positions_per_second": 19213.816658318035,
+      "required_pages_per_second_upper_estimate": 0.0040017313075081055
     },
     {
       "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": 28561.94370131757,
-      "required_pages_per_second_upper_estimate": 0.0017122449133311478
+      "required_token_positions_per_second": 28820.724987477053,
+      "required_pages_per_second_upper_estimate": 0.001727758456300264
     },
     {
       "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": 28561.94370131757,
-      "required_pages_per_second_upper_estimate": 0.002220565492064162
+      "required_token_positions_per_second": 28820.724987477053,
+      "required_pages_per_second_upper_estimate": 0.002240684598804479
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28561.94370131757,
-      "required_pages_per_second_upper_estimate": 0.36636194637469466
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28820.724987477053,
+      "required_pages_per_second_upper_estimate": 0.3696813148558517
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "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,
+    "producer_batch_seconds": 0.05214217556251555,
+    "producer_p95_seconds": 0.18714983300014865,
+    "gpu_update_seconds": 0.9593096638621472,
+    "producer_batches_per_second": 19.17833287951432,
+    "gpu_batches_per_second": 1.042416268354928,
+    "producer_headroom": 18.397960068083247,
     "sustainable": true,
-    "buffer_seconds": 60.016083566501976,
+    "buffer_seconds": 60.43650882331527,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 155,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 582705.1219094099,
-    "upload_bytes_per_second": 2783607.079390321,
+    "generation_bytes_per_second": 600660.1875822922,
+    "upload_bytes_per_second": 2907434.638492893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.51747724037992,
+    "estimated_full_shards_per_hour": 64.44384680081166,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 6396111,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.8616032419995463,
+      "bytes": 11914182,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -69,3 +69,4 @@
 {"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}
+{"step": 692200, "loss": 0.5909315943717957, "lr": 8e-05, "grad_norm": 0.21854433595812459, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101319637046, "tokens_per_second": 153310.22803827678, "data_wait_s": 1.6062000213423744e-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:f1bb3c3a2d82a0928c7175122b374e06cf0ac99bb8ab399adfc42d546b1467ac
-size 151326
+oid sha256:8ae8502675aae6e5f06d89d7f4ec8a5dcffd0a91487c5544fda20bd0f5074c2a
+size 153444