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": 77285.09116302131,
-      "required_pages_per_second_upper_estimate": 0.0053467619327756295
+      "required_token_positions_per_second": 75302.37762640312,
+      "required_pages_per_second_upper_estimate": 0.005209593209783149
     },
     {
       "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": 9660.636395377664,
-      "required_pages_per_second_upper_estimate": 0.00404470672171641
+      "required_token_positions_per_second": 9412.79720330039,
+      "required_pages_per_second_upper_estimate": 0.003940941627464508
     },
     {
       "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": 19321.272790755327,
-      "required_pages_per_second_upper_estimate": 0.004024111586086015
+      "required_token_positions_per_second": 18825.59440660078,
+      "required_pages_per_second_upper_estimate": 0.003920874850584661
     },
     {
       "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": 28981.90918613299,
-      "required_pages_per_second_upper_estimate": 0.0017374212028956665
+      "required_token_positions_per_second": 28238.39160990117,
+      "required_pages_per_second_upper_estimate": 0.0016928484594861727
     },
     {
       "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": 28981.90918613299,
-      "required_pages_per_second_upper_estimate": 0.002253215961275615
+      "required_token_positions_per_second": 28238.39160990117,
+      "required_pages_per_second_upper_estimate": 0.0021954107401117836
     },
     {
       "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": 28981.90918613299,
-      "required_pages_per_second_upper_estimate": 0.37174881269010135
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28238.39160990117,
+      "required_pages_per_second_upper_estimate": 0.36221176754917417
     }
   ],
   "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": 0.07907406302346232,
-    "producer_p95_seconds": 0.16473104399938165,
-    "gpu_update_seconds": 0.9539744197814536,
-    "producer_batches_per_second": 12.646371790751246,
-    "gpu_batches_per_second": 1.0482461366512221,
-    "producer_headroom": 12.064315191422462,
+    "producer_batch_seconds": 0.4228311505234714,
+    "producer_p95_seconds": 0.1703095390003,
+    "gpu_update_seconds": 0.9790925907517282,
+    "producer_batches_per_second": 2.365010238157678,
+    "gpu_batches_per_second": 1.0213538632053374,
+    "producer_headroom": 2.315564001232162,
     "sustainable": true,
-    "buffer_seconds": 60.10038844623158,
+    "buffer_seconds": 60.70374062660714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 505567.45617399353,
-    "upload_bytes_per_second": 4235740.382534904,
+    "generation_bytes_per_second": 449538.26010849635,
+    "upload_bytes_per_second": 4287840.198226729,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.48300708689551,
+    "estimated_full_shards_per_hour": 96.46044590417068,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 7420221,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.6006398999998055,
+      "bytes": 9959535,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 847200, "loss": 0.5169481039047241, "lr": 3e-05, "grad_norm": 0.29016980662491804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124006924161, "tokens_per_second": 153178.67551990322, "data_wait_s": 1.0660999578249175e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 847300, "loss": 0.540375828742981, "lr": 3e-05, "grad_norm": 0.3112236899748545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124021577613, "tokens_per_second": 152124.6183945812, "data_wait_s": 1.5653999980713706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 847400, "loss": 0.5843215584754944, "lr": 3e-05, "grad_norm": 0.34049848095453644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124036217268, "tokens_per_second": 152833.80106506718, "data_wait_s": 8.40600023366278e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 847500, "loss": 0.5078483819961548, "lr": 3e-05, "grad_norm": 0.29790460234382193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124050872630, "tokens_per_second": 151769.51493085473, "data_wait_s": 9.835000128077809e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c99d7dc19503e50b78890abbe2eb36104ade708b8731e51eae458d1adcbad5bc
-size 104730
+oid sha256:27c85c06a8c7f191e832f804bf6fb928adb748e38a0055eed12b7cc63f730d64
+size 106848