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": 76924.61604311105,
-      "required_pages_per_second_upper_estimate": 0.005321823427562724
+      "required_token_positions_per_second": 77457.37587181755,
+      "required_pages_per_second_upper_estimate": 0.005358680988685754
     },
     {
       "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": 9615.577005388881,
-      "required_pages_per_second_upper_estimate": 0.004025841296075168
+      "required_token_positions_per_second": 9682.171983977194,
+      "required_pages_per_second_upper_estimate": 0.004053723222948805
     },
     {
       "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": 19231.154010777762,
-      "required_pages_per_second_upper_estimate": 0.004005342220808735
+      "required_token_positions_per_second": 19364.343967954388,
+      "required_pages_per_second_upper_estimate": 0.004033082176433253
     },
     {
       "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": 28846.731016166643,
-      "required_pages_per_second_upper_estimate": 0.0017293174780112983
+      "required_token_positions_per_second": 29046.51595193158,
+      "required_pages_per_second_upper_estimate": 0.0017412942798564724
     },
     {
       "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": 28846.731016166643,
-      "required_pages_per_second_upper_estimate": 0.002242706453146663
+      "required_token_positions_per_second": 29046.51595193158,
+      "required_pages_per_second_upper_estimate": 0.002258238853141325
     },
     {
       "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": 28846.731016166643,
-      "required_pages_per_second_upper_estimate": 0.37001489226878365
+      "required_token_positions_per_second": 29046.51595193158,
+      "required_pages_per_second_upper_estimate": 0.3725775189124252
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.18576636082036657,
-    "producer_p95_seconds": 0.27227727900026366,
-    "gpu_update_seconds": 0.9584448228988292,
-    "producer_batches_per_second": 5.38310593793128,
-    "gpu_batches_per_second": 1.0433568799250088,
-    "producer_headroom": 5.159410017326182,
+    "producer_batch_seconds": 0.12092556748442007,
+    "producer_p95_seconds": 0.299332721002429,
+    "gpu_update_seconds": 0.9518525404476753,
+    "producer_batches_per_second": 8.269549780106171,
+    "gpu_batches_per_second": 1.0505828975669698,
+    "producer_headroom": 7.871391966552574,
     "sustainable": true,
-    "buffer_seconds": 60.38202384262624,
+    "buffer_seconds": 60.91856258865122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 529005.7333547273,
-    "upload_bytes_per_second": 5313811.862840319,
+    "generation_bytes_per_second": 527644.8579318648,
+    "upload_bytes_per_second": 5437383.162614947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.51231575471272,
+    "estimated_full_shards_per_hour": 113.22030356852491,
     "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": 1.6433775559999049,
-      "bytes": 5520234,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.1214206500008004,
       "bytes": 6053975,
@@ -562,6 +556,12 @@
       "bytes": 9800135,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.239405350999732,
+      "bytes": 13306132,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -244,3 +244,4 @@
 {"step": 669400, "loss": 0.5363394021987915, "lr": 0.0001, "grad_norm": 0.2073551345685677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97983400795, "tokens_per_second": 151838.5984416247, "data_wait_s": 1.1027001164620742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669500, "loss": 0.5883676409721375, "lr": 0.0001, "grad_norm": 0.18916531821171728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97997909063, "tokens_per_second": 151322.09718859673, "data_wait_s": 1.551699824631214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669600, "loss": 0.5544153451919556, "lr": 0.0001, "grad_norm": 0.2647767470857639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98012503363, "tokens_per_second": 152129.6504761414, "data_wait_s": 2.0611001673387364e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 669700, "loss": 0.5682086944580078, "lr": 0.0001, "grad_norm": 0.1946624194911147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98027024485, "tokens_per_second": 150728.8789047591, "data_wait_s": 9.079998562810943e-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 sha256:031fd9287eb3ff4b539d6d56cb7b9c01be1be2459da0d381d82dcec41f6a0639
-size 524325
+oid sha256:57dd6aca94443dee1c4beab7f6eabe4ab702248e410ced5c647613d80f10e12f
+size 526443