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": 77278.22278297294,
-      "required_pages_per_second_upper_estimate": 0.005346286762307046
+      "required_token_positions_per_second": 77270.83335266839,
+      "required_pages_per_second_upper_estimate": 0.005345775544372737
     },
     {
       "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": 9659.777847871617,
-      "required_pages_per_second_upper_estimate": 0.004044347265804139
+      "required_token_positions_per_second": 9658.854169083548,
+      "required_pages_per_second_upper_estimate": 0.004043960540784181
     },
     {
       "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": 19319.555695743235,
-      "required_pages_per_second_upper_estimate": 0.004023753960477837
+      "required_token_positions_per_second": 19317.708338167096,
+      "required_pages_per_second_upper_estimate": 0.004023369204612823
     },
     {
       "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": 28979.33354361485,
-      "required_pages_per_second_upper_estimate": 0.0017372667970594825
+      "required_token_positions_per_second": 28976.562507250645,
+      "required_pages_per_second_upper_estimate": 0.0017371006776605758
     },
     {
       "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": 28979.33354361485,
-      "required_pages_per_second_upper_estimate": 0.002253015716398884
+      "required_token_positions_per_second": 28976.562507250645,
+      "required_pages_per_second_upper_estimate": 0.0022528002805100675
     },
     {
       "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": 28979.33354361485,
-      "required_pages_per_second_upper_estimate": 0.3717157751133881
+      "required_token_positions_per_second": 28976.562507250645,
+      "required_pages_per_second_upper_estimate": 0.3716802312342151
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.29497392322659266,
-    "producer_p95_seconds": 0.3463655170016864,
-    "gpu_update_seconds": 0.9540592076898128,
-    "producer_batches_per_second": 3.390130181886693,
-    "gpu_batches_per_second": 1.0481529782846808,
-    "producer_headroom": 3.2343849152961393,
+    "producer_batch_seconds": 0.08491894143762124,
+    "producer_p95_seconds": 0.31772341200121446,
+    "gpu_update_seconds": 0.9541504446250929,
+    "producer_batches_per_second": 11.775935769695955,
+    "gpu_batches_per_second": 1.0480527527217391,
+    "producer_headroom": 11.236014350531931,
     "sustainable": true,
-    "buffer_seconds": 60.10573008445821,
+    "buffer_seconds": 60.111478011380854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 163,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 604429.2613845476,
-    "upload_bytes_per_second": 6028483.943791279,
+    "generation_bytes_per_second": 594530.0969969413,
+    "upload_bytes_per_second": 6027298.662325634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.84822455001984,
+    "estimated_full_shards_per_hour": 63.78615943160619,
     "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": 1.1336718419988756,
-      "bytes": 6075801,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0171541380004783,
       "bytes": 7685246,
@@ -562,6 +556,12 @@
       "bytes": 8446837,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.1454860439989716,
+      "bytes": 6061904,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -306,3 +306,4 @@
 {"step": 515600, "loss": 0.5998923778533936, "lr": 0.0003, "grad_norm": 0.16084814012242907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75466232709, "tokens_per_second": 151145.74336640813, "data_wait_s": 1.04649989225436e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515700, "loss": 0.5841395258903503, "lr": 0.0003, "grad_norm": 0.1735427580705892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75480932606, "tokens_per_second": 153203.58819828866, "data_wait_s": 1.6792997485026717e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 515800, "loss": 0.5726343393325806, "lr": 0.0003, "grad_norm": 0.1732284085563092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75495617983, "tokens_per_second": 153038.2169160619, "data_wait_s": 1.7416001355741173e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 515900, "loss": 0.6434698104858398, "lr": 0.0003, "grad_norm": 0.1668728827510193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75510359290, "tokens_per_second": 153016.33690597114, "data_wait_s": 9.945004421751946e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56d48f9ad494c129fa7b223770619d6bfa94d7883a625c719510b0c144f159ef
-size 657207
+oid sha256:07ce4887841a4f63509428759be2dff4b5fda89a574a27b2e10e401b194fdf78
+size 659325