Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74804.07113813485,
-      "required_pages_per_second_upper_estimate": 0.006483827297068538
+      "required_token_positions_per_second": 75826.98946208876,
+      "required_pages_per_second_upper_estimate": 0.006572491264826073
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9350.508892266856,
-      "required_pages_per_second_upper_estimate": 0.0037300577996915813
+      "required_token_positions_per_second": 9478.373682761096,
+      "required_pages_per_second_upper_estimate": 0.0037810649763687153
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18701.017784533713,
-      "required_pages_per_second_upper_estimate": 0.004326151337779938
+      "required_token_positions_per_second": 18956.74736552219,
+      "required_pages_per_second_upper_estimate": 0.004385309875654715
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28051.52667680057,
-      "required_pages_per_second_upper_estimate": 0.001708190799630964
+      "required_token_positions_per_second": 28435.121048283287,
+      "required_pages_per_second_upper_estimate": 0.0017315496842901332
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28051.52667680057,
-      "required_pages_per_second_upper_estimate": 0.0021594841751758726
+      "required_token_positions_per_second": 28435.121048283287,
+      "required_pages_per_second_upper_estimate": 0.0021890143317497943
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28051.52667680057,
-      "required_pages_per_second_upper_estimate": 0.3483060789053549
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28435.121048283287,
+      "required_pages_per_second_upper_estimate": 0.3530690371913939
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09430267088316668,
-    "producer_p95_seconds": 0.37259805100620724,
-    "gpu_update_seconds": 0.9856148051601662,
-    "producer_batches_per_second": 10.604153526456514,
-    "gpu_batches_per_second": 1.0145951489004836,
-    "producer_headroom": 10.451610711866927,
+    "producer_batch_seconds": 0.11475334957049199,
+    "producer_p95_seconds": 0.3841020199979539,
+    "gpu_update_seconds": 0.9723187023910768,
+    "producer_batches_per_second": 8.714342576864901,
+    "gpu_batches_per_second": 1.0284693666190425,
+    "producer_headroom": 8.473118266528592,
     "sustainable": true,
-    "buffer_seconds": 60.12250311477014,
+    "buffer_seconds": 60.283759548246756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430249.9474809038,
-    "upload_bytes_per_second": 4228944.919471442,
+    "generation_bytes_per_second": 428128.65490619396,
+    "upload_bytes_per_second": 4230360.3492596345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.32162302322708,
+    "estimated_full_shards_per_hour": 91.86644301785816,
     "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": 1.03730981300032,
-      "bytes": 6215262,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9270265450031729,
       "bytes": 1100519,
@@ -562,6 +556,12 @@
       "bytes": 7551543,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9091981460005627,
+      "bytes": 5814772,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -340,3 +340,4 @@
 {"step": 34000, "loss": 0.6324362829327583, "lr": 0.0003, "grad_norm": 0.1366603905928768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4975444748, "tokens_per_second": 148308.1229782295, "data_wait_s": 1.0525000107008964e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34100, "loss": 0.6961428821086884, "lr": 0.0003, "grad_norm": 0.13530116001377898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4989998423, "tokens_per_second": 148800.8387634716, "data_wait_s": 9.483999747317284e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 34200, "loss": 0.6966866441071033, "lr": 0.0003, "grad_norm": 0.14291184913415805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5004702933, "tokens_per_second": 149746.034129097, "data_wait_s": 1.497899938840419e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 34300, "loss": 0.5933354459702969, "lr": 0.0003, "grad_norm": 0.15678194327714132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5019240778, "tokens_per_second": 147608.11027876544, "data_wait_s": 1.2222000805195421e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8061ac20428b46c5dbe97ccb5fc5c2147fd2424869ad38257e8406d61489a614
-size 723445
+oid sha256:80c19c24ba17f08b375b1f6bb1fdeba2f42ee38dd6fd96942fad4df7b0c57d35
+size 725563