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": 77101.01904508268,
-      "required_pages_per_second_upper_estimate": 0.005334027396550478
+      "required_token_positions_per_second": 75042.20393901474,
+      "required_pages_per_second_upper_estimate": 0.005191593790403495
     },
     {
       "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": 9637.627380635335,
-      "required_pages_per_second_upper_estimate": 0.004035073327726662
+      "required_token_positions_per_second": 9380.275492376843,
+      "required_pages_per_second_upper_estimate": 0.00392732546623137
     },
     {
       "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": 19275.25476127067,
-      "required_pages_per_second_upper_estimate": 0.004014527244121408
+      "required_token_positions_per_second": 18760.550984753685,
+      "required_pages_per_second_upper_estimate": 0.003907328021124296
     },
     {
       "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": 28912.882141906004,
-      "required_pages_per_second_upper_estimate": 0.0017332831369924525
+      "required_token_positions_per_second": 28140.826477130526,
+      "required_pages_per_second_upper_estimate": 0.0016869995787499043
     },
     {
       "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": 28912.882141906004,
-      "required_pages_per_second_upper_estimate": 0.0022478494122048468
+      "required_token_positions_per_second": 28140.826477130526,
+      "required_pages_per_second_upper_estimate": 0.0021878254801825318
     },
     {
       "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": 28912.882141906004,
-      "required_pages_per_second_upper_estimate": 0.3708634078822232
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28140.826477130526,
+      "required_pages_per_second_upper_estimate": 0.3609603067832702
     }
   ],
   "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": 0.13824038977321607,
-    "producer_p95_seconds": 0.3291892720008036,
-    "gpu_update_seconds": 0.9562519524792481,
-    "producer_batches_per_second": 7.233775900375456,
-    "gpu_batches_per_second": 1.0457494987668547,
-    "producer_headroom": 6.91731232853136,
+    "producer_batch_seconds": 0.3391172111871015,
+    "producer_p95_seconds": 0.355980565000209,
+    "gpu_update_seconds": 0.9824871356379303,
+    "producer_batches_per_second": 2.948832931538438,
+    "gpu_batches_per_second": 1.0178250317249178,
+    "producer_headroom": 2.897190420382001,
     "sustainable": true,
-    "buffer_seconds": 60.24387300619263,
+    "buffer_seconds": 60.91420240955168,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406148.286174199,
-    "upload_bytes_per_second": 4917912.886723278,
+    "generation_bytes_per_second": 385732.1530428105,
+    "upload_bytes_per_second": 4949797.310566265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.14996756476859,
+    "estimated_full_shards_per_hour": 82.769140658028,
     "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": 0.8581832199997734,
-      "bytes": 2650389,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.6456713849984226,
       "bytes": 6071875,
@@ -562,6 +556,12 @@
       "bytes": 1252954,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2454847960034385,
+      "bytes": 7379042,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -397,3 +397,4 @@
 {"step": 684700, "loss": 0.59892338514328, "lr": 0.0001, "grad_norm": 0.2340129714638774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100222432027, "tokens_per_second": 149699.38474605192, "data_wait_s": 1.073999737855047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684800, "loss": 0.565108060836792, "lr": 0.0001, "grad_norm": 0.22957951672422106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100237080110, "tokens_per_second": 152459.09059993987, "data_wait_s": 9.927003702614456e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 684900, "loss": 0.5829205513000488, "lr": 0.0001, "grad_norm": 0.2044919994639403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100251771725, "tokens_per_second": 152254.12713306598, "data_wait_s": 1.198000245494768e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685000, "loss": 0.5596920847892761, "lr": 0.0001, "grad_norm": 0.2115820071756629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100266499039, "tokens_per_second": 153899.60702044988, "data_wait_s": 8.930001058615744e-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 sha256:43be9cb899b44abae5e41055a55d52918d8cbb467fb44c7e45c352e25cb57653
-size 850728
+oid sha256:415103dd0bc580394fd3de86ff784b135ec1c1fefbca12cbbc054172e79bd077
+size 852846