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": 76753.75015337917,
-      "required_pages_per_second_upper_estimate": 0.005310002528847586
+      "required_token_positions_per_second": 75334.04829189886,
+      "required_pages_per_second_upper_estimate": 0.005211784259908213
     },
     {
       "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": 9594.218769172396,
-      "required_pages_per_second_upper_estimate": 0.004016899048582016
+      "required_token_positions_per_second": 9416.756036487357,
+      "required_pages_per_second_upper_estimate": 0.003942599108250054
     },
     {
       "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": 19188.43753834479,
-      "required_pages_per_second_upper_estimate": 0.00399644550611008
+      "required_token_positions_per_second": 18833.512072974714,
+      "required_pages_per_second_upper_estimate": 0.0039225238916874485
     },
     {
       "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": 28782.656307517187,
-      "required_pages_per_second_upper_estimate": 0.0017254762970641787
+      "required_token_positions_per_second": 28250.268109462068,
+      "required_pages_per_second_upper_estimate": 0.0016935604375004835
     },
     {
       "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": 28782.656307517187,
-      "required_pages_per_second_upper_estimate": 0.0022377249263840263
+      "required_token_positions_per_second": 28250.268109462068,
+      "required_pages_per_second_upper_estimate": 0.0021963340857134423
     },
     {
       "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": 28782.656307517187,
-      "required_pages_per_second_upper_estimate": 0.36919301070429045
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28250.268109462068,
+      "required_pages_per_second_upper_estimate": 0.36236410653354967
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.050036518242251304,
-    "producer_p95_seconds": 0.17040360800001508,
-    "gpu_update_seconds": 0.9605784714449428,
-    "producer_batches_per_second": 19.98540336396929,
-    "gpu_batches_per_second": 1.0410393629744352,
-    "producer_headroom": 19.19754821457224,
+    "producer_batch_seconds": 0.11343236420318448,
+    "producer_p95_seconds": 0.1876597070004209,
+    "gpu_update_seconds": 0.9786809772166252,
+    "producer_batches_per_second": 8.815826127089805,
+    "gpu_batches_per_second": 1.0217834240980204,
+    "producer_headroom": 8.627881329032107,
     "sustainable": true,
-    "buffer_seconds": 60.5164437010314,
+    "buffer_seconds": 60.67822058743076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381899.42008505826,
-    "upload_bytes_per_second": 3929743.295543704,
+    "generation_bytes_per_second": 396249.2120486046,
+    "upload_bytes_per_second": 3739854.0668771025,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.94672538675128,
+    "estimated_full_shards_per_hour": 85.02585669606785,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 10022494,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.460268009000174,
+      "bytes": 1130080,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 805600, "loss": 0.5071325898170471, "lr": 3e-05, "grad_norm": 0.27077951718871107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117916661451, "tokens_per_second": 151694.02694097103, "data_wait_s": 2.65179996858933e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805700, "loss": 0.5595177412033081, "lr": 3e-05, "grad_norm": 0.27615446483562395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117931176621, "tokens_per_second": 150598.8711746715, "data_wait_s": 1.2598000466823578e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805800, "loss": 0.5090294480323792, "lr": 3e-05, "grad_norm": 0.26825289216626286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117945672353, "tokens_per_second": 149602.61917522838, "data_wait_s": 1.5653999980713706e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 805900, "loss": 0.5663489699363708, "lr": 3e-05, "grad_norm": 0.2863788143839216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117960273810, "tokens_per_second": 151720.84326337787, "data_wait_s": 1.4418999853660353e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:204e7d20fb796178fe24776313d74aa1aa470f6f44c22ee1e526c8aeb50cb25d
-size 123792
+oid sha256:223baff209fd6fefd3114b2ef75458d719504361e929f893f76ef3d802b191bf
+size 125910