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": 75768.95887930205,
-      "required_pages_per_second_upper_estimate": 0.00524187238347636
+      "required_token_positions_per_second": 77425.99555809847,
+      "required_pages_per_second_upper_estimate": 0.005356510025770309
     },
     {
       "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": 9471.119859912757,
-      "required_pages_per_second_upper_estimate": 0.003965360105872543
+      "required_token_positions_per_second": 9678.249444762308,
+      "required_pages_per_second_upper_estimate": 0.004052080937691466
     },
     {
       "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": 18942.239719825513,
-      "required_pages_per_second_upper_estimate": 0.003945168993185602
+      "required_token_positions_per_second": 19356.498889524617,
+      "required_pages_per_second_upper_estimate": 0.004031448253484963
     },
     {
       "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": 28413.359579738273,
-      "required_pages_per_second_upper_estimate": 0.0017033375221172861
+      "required_token_positions_per_second": 29034.748334286927,
+      "required_pages_per_second_upper_estimate": 0.0017405888291467874
     },
     {
       "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": 28413.359579738273,
-      "required_pages_per_second_upper_estimate": 0.002209013730163853
+      "required_token_positions_per_second": 29034.748334286927,
+      "required_pages_per_second_upper_estimate": 0.002257323972629733
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28413.359579738273,
-      "required_pages_per_second_upper_estimate": 0.3644560688002754
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29034.748334286927,
+      "required_pages_per_second_upper_estimate": 0.372426576548363
     }
   ],
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0928057924377299,
-    "producer_p95_seconds": 0.330546819997835,
-    "gpu_update_seconds": 0.9730633902129597,
-    "producer_batches_per_second": 10.775189497691882,
-    "gpu_batches_per_second": 1.0276822764662281,
-    "producer_headroom": 10.484942422811141,
+    "producer_batch_seconds": 0.1494182022734094,
+    "producer_p95_seconds": 0.328610063996166,
+    "gpu_update_seconds": 0.9522383208450501,
+    "producer_batches_per_second": 6.692625026836914,
+    "gpu_batches_per_second": 1.0501572748222991,
+    "producer_headroom": 6.372974017600741,
     "sustainable": true,
-    "buffer_seconds": 60.3299301932035,
+    "buffer_seconds": 60.94325253408321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281762.4828313415,
-    "upload_bytes_per_second": 3620895.68875037,
+    "generation_bytes_per_second": 283043.6758186193,
+    "upload_bytes_per_second": 3586049.730437874,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.459669720699154,
+    "estimated_full_shards_per_hour": 60.734583911122655,
     "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": 1.681274881997524,
-      "bytes": 6344982,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.540335444999073,
       "bytes": 2545551,
@@ -562,6 +556,12 @@
       "bytes": 5919524,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.4157910470021307,
+      "bytes": 1321363,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -390,3 +390,4 @@
 {"step": 881500, "loss": 0.537477433681488, "lr": 3e-05, "grad_norm": 0.31144380978314007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129027006398, "tokens_per_second": 153477.819587664, "data_wait_s": 1.7523998394608498e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881600, "loss": 0.49659985303878784, "lr": 3e-05, "grad_norm": 0.30103455746506924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129041759699, "tokens_per_second": 153413.5448973277, "data_wait_s": 1.6828002117108554e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881700, "loss": 0.5240030884742737, "lr": 3e-05, "grad_norm": 0.3224385091833834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129056299726, "tokens_per_second": 151102.26517850452, "data_wait_s": 1.588300074217841e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 881800, "loss": 0.5664958953857422, "lr": 3e-05, "grad_norm": 0.32144307865781946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129070716406, "tokens_per_second": 149217.93866730624, "data_wait_s": 1.19440010166727e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:829d78a8d09a4cfbb685366970697b3639db2936f78b7746d427715f6c4e876a
-size 836685
+oid sha256:21ab1342625af411a5b5f976881c5d95878ce5ea1d495d64686875f3c9ea440c
+size 838803