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": 76954.13948611787,
-      "required_pages_per_second_upper_estimate": 0.005323865927853761
+      "required_token_positions_per_second": 77501.7127357748,
+      "required_pages_per_second_upper_estimate": 0.005361748315810014
     },
     {
       "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": 9619.267435764734,
-      "required_pages_per_second_upper_estimate": 0.0040273864022085885
+      "required_token_positions_per_second": 9687.71409197185,
+      "required_pages_per_second_upper_estimate": 0.004056043587833792
     },
     {
       "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": 19238.534871529468,
-      "required_pages_per_second_upper_estimate": 0.004006879459456928
+      "required_token_positions_per_second": 19375.4281839437,
+      "required_pages_per_second_upper_estimate": 0.004035390726313394
     },
     {
       "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": 28857.8023072942,
-      "required_pages_per_second_upper_estimate": 0.0017299811850095127
+      "required_token_positions_per_second": 29063.14227591555,
+      "required_pages_per_second_upper_estimate": 0.0017422910025923848
     },
     {
       "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": 28857.8023072942,
-      "required_pages_per_second_upper_estimate": 0.0022435671973343677
+      "required_token_positions_per_second": 29063.14227591555,
+      "required_pages_per_second_upper_estimate": 0.002259531476699608
     },
     {
       "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": 28857.8023072942,
-      "required_pages_per_second_upper_estimate": 0.3701569029039417
+      "required_token_positions_per_second": 29063.14227591555,
+      "required_pages_per_second_upper_estimate": 0.3727907835445357
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.45304742894534655,
-    "producer_p95_seconds": 0.18793714299954445,
-    "gpu_update_seconds": 0.9580771156995415,
-    "producer_batches_per_second": 2.207274417885804,
-    "gpu_batches_per_second": 1.0437573172487775,
-    "producer_headroom": 2.1147391078454154,
+    "producer_batch_seconds": 0.0941240886406689,
+    "producer_p95_seconds": 0.19431053499829432,
+    "gpu_update_seconds": 0.9513080085256896,
+    "producer_batches_per_second": 10.624272855566566,
+    "gpu_batches_per_second": 1.051184254771251,
+    "producer_headroom": 10.106955852262573,
     "sustainable": true,
-    "buffer_seconds": 60.358858289071115,
+    "buffer_seconds": 60.88371254564414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328459.3244965244,
-    "upload_bytes_per_second": 2948605.3177353926,
+    "generation_bytes_per_second": 354827.8560342017,
+    "upload_bytes_per_second": 3000828.495707168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.47972489520835,
+    "estimated_full_shards_per_hour": 76.13779793519534,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 6431657,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.255959656999039,
+      "bytes": 8183161,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 652900, "loss": 0.5208961963653564, "lr": 0.0001, "grad_norm": 0.17214884091978144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95567013116, "tokens_per_second": 153149.21776560624, "data_wait_s": 1.3562000276579056e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653000, "loss": 0.5420287251472473, "lr": 0.0001, "grad_norm": 0.18063351481286485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95581694897, "tokens_per_second": 153296.25834348908, "data_wait_s": 7.841999831725843e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 653100, "loss": 0.6101648807525635, "lr": 0.0001, "grad_norm": 0.20305761003838585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95596603474, "tokens_per_second": 156076.9113091464, "data_wait_s": 9.385001249029301e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 653200, "loss": 0.5545850396156311, "lr": 0.0001, "grad_norm": 0.21726353145859434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95611313773, "tokens_per_second": 153295.92390513566, "data_wait_s": 7.992999599082395e-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 sha256:5a7e97899c8c449b32e955bf34f0ae175637496893de2f4986ec1e6db5ecba09
-size 172506
+oid sha256:bdad574220cde11bd3b32a64004ce4cde8ef98829b7005e5ec083a856761b352
+size 174624