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": 76879.8667989697,
-      "required_pages_per_second_upper_estimate": 0.005318727571020474
+      "required_token_positions_per_second": 74097.23748140519,
+      "required_pages_per_second_upper_estimate": 0.0051262188182418
     },
     {
       "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": 9609.983349871212,
-      "required_pages_per_second_upper_estimate": 0.004023499349318731
+      "required_token_positions_per_second": 9262.154685175648,
+      "required_pages_per_second_upper_estimate": 0.0038778707508991756
     },
     {
       "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": 19219.966699742425,
-      "required_pages_per_second_upper_estimate": 0.004003012198949306
+      "required_token_positions_per_second": 18524.309370351297,
+      "required_pages_per_second_upper_estimate": 0.0038581251229546066
     },
     {
       "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": 28829.950049613635,
-      "required_pages_per_second_upper_estimate": 0.0017283114846894977
+      "required_token_positions_per_second": 27786.464055526947,
+      "required_pages_per_second_upper_estimate": 0.0016657560926548582
     },
     {
       "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": 28829.950049613635,
-      "required_pages_per_second_upper_estimate": 0.002241401806808838
+      "required_token_positions_per_second": 27786.464055526947,
+      "required_pages_per_second_upper_estimate": 0.00216027536057842
     },
     {
       "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": 28829.950049613635,
-      "required_pages_per_second_upper_estimate": 0.36979964404783977
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 27786.464055526947,
+      "required_pages_per_second_upper_estimate": 0.3564149261236637
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09720631545326341,
-    "producer_p95_seconds": 0.2958814740013622,
-    "gpu_update_seconds": 0.9590027021351195,
-    "producer_batches_per_second": 10.287397432328333,
-    "gpu_batches_per_second": 1.0427499294565117,
-    "producer_headroom": 9.865641935540761,
+    "producer_batch_seconds": 0.39705784404699784,
+    "producer_p95_seconds": 0.3115077740003471,
+    "gpu_update_seconds": 0.9950168522612216,
+    "producer_batches_per_second": 2.518524731327647,
+    "gpu_batches_per_second": 1.005008103860205,
+    "producer_headroom": 2.5059745505076743,
     "sustainable": true,
-    "buffer_seconds": 60.41717023451253,
+    "buffer_seconds": 60.69602798793452,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315228.7005192869,
-    "upload_bytes_per_second": 5604018.007299239,
+    "generation_bytes_per_second": 306672.75069183187,
+    "upload_bytes_per_second": 5598761.825171184,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.64074098285633,
+    "estimated_full_shards_per_hour": 65.80483332220284,
     "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": 1.0596161510002275,
-      "bytes": 5840926,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9319052459977684,
       "bytes": 2836860,
@@ -562,6 +556,12 @@
       "bytes": 2427794,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1882967479978106,
+      "bytes": 6156483,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -314,3 +314,4 @@
 {"step": 676400, "loss": 0.5948325395584106, "lr": 0.0001, "grad_norm": 0.20703745292978173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99007715710, "tokens_per_second": 153685.92040189268, "data_wait_s": 1.4428998838411644e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676500, "loss": 0.5490423440933228, "lr": 0.0001, "grad_norm": 0.20256718652373829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99022276854, "tokens_per_second": 151011.49398025637, "data_wait_s": 6.705999112455174e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 676600, "loss": 0.5414052605628967, "lr": 0.0001, "grad_norm": 0.19344810540610674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99036849852, "tokens_per_second": 151399.6458544073, "data_wait_s": 1.0105999535880983e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 676700, "loss": 0.5441694855690002, "lr": 0.0001, "grad_norm": 0.18920825808135824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99051399436, "tokens_per_second": 151587.82029818394, "data_wait_s": 1.758299913490191e-05, "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 sha256:571b85664f5a152e57741cf751baf997f8cdde94e8ac95e9e013b29d740ff2ab
-size 674151
+oid sha256:b6fcc922de8231de589f223780e9473216a197b1794b1257145a031e70cc559c
+size 676269