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": 76458.31864812189,
-      "required_pages_per_second_upper_estimate": 0.005289563891818352
+      "required_token_positions_per_second": 75529.13203838936,
+      "required_pages_per_second_upper_estimate": 0.005225280606147082
     },
     {
       "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": 9557.289831015236,
-      "required_pages_per_second_upper_estimate": 0.0040014376733395335
+      "required_token_positions_per_second": 9441.14150479867,
+      "required_pages_per_second_upper_estimate": 0.003952808794605516
     },
     {
       "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": 19114.57966203047,
-      "required_pages_per_second_upper_estimate": 0.003981062858236987
+      "required_token_positions_per_second": 18882.28300959734,
+      "required_pages_per_second_upper_estimate": 0.003932681591609851
     },
     {
       "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": 28671.86949304571,
-      "required_pages_per_second_upper_estimate": 0.0017188347966982844
+      "required_token_positions_per_second": 28323.424514396007,
+      "required_pages_per_second_upper_estimate": 0.0016979460522729107
     },
     {
       "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": 28671.86949304571,
-      "required_pages_per_second_upper_estimate": 0.002229111738858566
+      "required_token_positions_per_second": 28323.424514396007,
+      "required_pages_per_second_upper_estimate": 0.002202021674416038
     },
     {
       "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": 28671.86949304571,
-      "required_pages_per_second_upper_estimate": 0.3677719564018639
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28323.424514396007,
+      "required_pages_per_second_upper_estimate": 0.3633024783468145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.25890131135165007,
-    "producer_p95_seconds": 0.26839650299734785,
-    "gpu_update_seconds": 0.9642901034655572,
-    "producer_batches_per_second": 3.8624756080967093,
-    "gpu_batches_per_second": 1.037032316733424,
-    "producer_headroom": 3.7245470037647666,
+    "producer_batch_seconds": 0.11752881725783482,
+    "producer_p95_seconds": 0.29009248399961507,
+    "gpu_update_seconds": 0.9761531479340463,
+    "producer_batches_per_second": 8.508551547882925,
+    "gpu_batches_per_second": 1.0244294167533279,
+    "producer_headroom": 8.30564937782502,
     "sustainable": true,
-    "buffer_seconds": 60.750276518330104,
+    "buffer_seconds": 60.52149517191087,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443450.4259182352,
-    "upload_bytes_per_second": 5344720.103505357,
+    "generation_bytes_per_second": 459001.83914572606,
+    "upload_bytes_per_second": 5447226.013311381,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.1541384044675,
+    "estimated_full_shards_per_hour": 98.49110966471515,
     "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": 0.9337214369988942,
-      "bytes": 1171128,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.1392940650002856,
       "bytes": 6258107,
@@ -562,6 +556,12 @@
       "bytes": 6164100,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3281419349987118,
+      "bytes": 11547797,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 672600, "loss": 0.4951384961605072, "lr": 0.0001, "grad_norm": 0.21271789587637635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98451413192, "tokens_per_second": 153848.4484991045, "data_wait_s": 7.1309987106360495e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 672700, "loss": 0.5589165091514587, "lr": 0.0001, "grad_norm": 0.23221826338201942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98465987188, "tokens_per_second": 151485.6750194696, "data_wait_s": 1.2708998838206753e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 672800, "loss": 0.5490952730178833, "lr": 0.0001, "grad_norm": 0.2002931100371676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98480494479, "tokens_per_second": 150418.87794062702, "data_wait_s": 8.977996913017705e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 672900, "loss": 0.5461544394493103, "lr": 0.0001, "grad_norm": 0.19277019882890248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98495069753, "tokens_per_second": 152051.37842661559, "data_wait_s": 9.282000974053517e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afca74707381fed2d12ac204db1130a47a0b02003d49999f89fd735cc13518d8
-size 592884
+oid sha256:f736e696899d8b1a6fdaf56f3e6c42dd32facc036d80e26c4f744b0c3cebd820
+size 595002