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": 77438.48096486849,
-      "required_pages_per_second_upper_estimate": 0.005357373795180793
+      "required_token_positions_per_second": 76400.10574915432,
+      "required_pages_per_second_upper_estimate": 0.005285536588395241
     },
     {
       "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": 9679.810120608561,
-      "required_pages_per_second_upper_estimate": 0.004052734360077674
+      "required_token_positions_per_second": 9550.01321864429,
+      "required_pages_per_second_upper_estimate": 0.003998391107692764
     },
     {
       "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": 19359.620241217122,
-      "required_pages_per_second_upper_estimate": 0.004032098348726932
+      "required_token_positions_per_second": 19100.02643728858,
+      "required_pages_per_second_upper_estimate": 0.003978031805317596
     },
     {
       "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": 29039.43036182568,
-      "required_pages_per_second_upper_estimate": 0.0017408695095486935
+      "required_token_positions_per_second": 28650.039655932866,
+      "required_pages_per_second_upper_estimate": 0.0017175261312956007
     },
     {
       "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": 29039.43036182568,
-      "required_pages_per_second_upper_estimate": 0.0022576879796768084
+      "required_token_positions_per_second": 28650.039655932866,
+      "required_pages_per_second_upper_estimate": 0.0022274145650423485
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29039.43036182568,
-      "required_pages_per_second_upper_estimate": 0.37248663257046066
+      "required_token_positions_per_second": 28650.039655932866,
+      "required_pages_per_second_upper_estimate": 0.3674919466904332
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08227566177336598,
+    "producer_batch_seconds": 0.20316245513275533,
     "producer_p95_seconds": 0.19625516599990078,
-    "gpu_update_seconds": 0.9520847914546281,
-    "producer_batches_per_second": 12.154262614800588,
-    "gpu_batches_per_second": 1.050326618989644,
-    "producer_headroom": 11.5718885868972,
+    "gpu_update_seconds": 0.9650248422701445,
+    "producer_batches_per_second": 4.922169302130927,
+    "gpu_batches_per_second": 1.0362427537591459,
+    "producer_headroom": 4.7500156544158445,
     "sustainable": true,
-    "buffer_seconds": 60.9334266530962,
+    "buffer_seconds": 60.7965650630191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278990.3774512255,
-    "upload_bytes_per_second": 3541760.4497238207,
+    "generation_bytes_per_second": 284926.56673938985,
+    "upload_bytes_per_second": 3534641.107922144,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.864840437436804,
+    "estimated_full_shards_per_hour": 61.138608471262664,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 6125796,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0003182039999956,
+      "bytes": 2721093,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -108,3 +108,4 @@
 {"step": 655800, "loss": 0.5330632925033569, "lr": 0.0001, "grad_norm": 0.1840939121120962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95991926899, "tokens_per_second": 154325.79141993396, "data_wait_s": 6.632999429712072e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 655900, "loss": 0.6637529730796814, "lr": 0.0001, "grad_norm": 0.24339154460259993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96006503135, "tokens_per_second": 152470.70479462782, "data_wait_s": 1.6501999198226258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656000, "loss": 0.6607984900474548, "lr": 0.0001, "grad_norm": 0.3911945317755853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96020983712, "tokens_per_second": 151787.83773761903, "data_wait_s": 1.0215999282081611e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 656100, "loss": 0.537526547908783, "lr": 0.0001, "grad_norm": 0.17528571990657263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96035457912, "tokens_per_second": 150510.09848665766, "data_wait_s": 1.6138999853865243e-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:f45a6138ebb1b85074bbe8559eabe0f7b87c4e3aace0882250205b25d366e0a5
-size 234711
+oid sha256:20a3416122d0c9823110a0a7861b84a4dc98271ac85a20a256c5402246d879a2
+size 236829